# Load packages
import numpy as np # The 'numpy' package is needed for matrix operations and calculations
import pandas as pd # The 'pandas' package helps us to import and manage data
import math as math # Math package for data cleaning
from scipy import stats # Import scipy.package for confidence intervals
from sklearn.preprocessing import StandardScaler # Import the package for standardizing data
import D47calib as D47c # Import the package for treating clumped isotope data by Daëron and Vermeesch (2023; https://github.com/mdaeron/D47calib)
import matplotlib.pyplot as plt # The 'matplotlib' package contains tools needed to plot our data and results
import seaborn as sns # The 'seaborn' package is used to make our plots look nicer (e.g. enable heatmaps)
import warnings # The 'warnings' package is used to suppress warnings that might occur during the calculations
%matplotlib inline

# Load model SAT data as prior and show data structure
Lutetian_SAT = pd.read_csv('Lutetian case/SAT_Individual_values.csv') # Load the data for this assignment into Python and in the Jupyter environment.
Lutetian_SAT.head()

# Load model SST data as prior and show data structure
Lutetian_SST = pd.read_csv('Lutetian case/SST_Individual_values.csv') # Load the data for this assignment into Python and in the Jupyter environment.
Lutetian_SST.head()

# Load model SSS data as prior and show data structure
Lutetian_SSS = pd.read_csv('Lutetian case/SSS_Individual_values.csv') # Load the data for this assignment into Python and in the Jupyter environment.
Lutetian_SSS.head()

# Load model precipitation data as prior and show data structure
Lutetian_prec = pd.read_csv('Lutetian case/Precipitation_Individual_values.csv') # Load the data for this assignment into Python and in the Jupyter environment.
Lutetian_prec.head()

# Add a column to number the rows within each month consecutively
Lutetian_SAT['Cell'] = Lutetian_SAT.groupby('Month').cumcount() + 1
Lutetian_SST['Cell'] = Lutetian_SST.groupby('Month').cumcount() + 1
Lutetian_SSS['Cell'] = Lutetian_SSS.groupby('Month').cumcount() + 1
Lutetian_prec['Cell'] = Lutetian_prec.groupby('Month').cumcount() + 1

# Pivot the datasets to create separate columns for each month with 2-letter abbreviations
month_abbreviations = ['ja', 'fb', 'mr', 'ar', 'my', 'jn', 'jl', 'ag', 'sp', 'ot', 'nv', 'dc']
Lutetian_SAT_wide = Lutetian_SAT.pivot(index='Cell', columns='Month', values='Temperature')
Lutetian_SAT_wide.columns = [month_abbreviations[col - 1] for col in Lutetian_SAT_wide.columns]
Lutetian_SST_wide = Lutetian_SST.pivot(index='Cell', columns='Month', values='SST')
Lutetian_SST_wide.columns = [month_abbreviations[col - 1] for col in Lutetian_SST_wide.columns]
Lutetian_SSS_wide = Lutetian_SSS.pivot(index='Cell', columns='Month', values='SSS')
Lutetian_SSS_wide.columns = [month_abbreviations[col - 1] for col in Lutetian_SSS_wide.columns]
Lutetian_prec_wide = Lutetian_prec.pivot(index='Cell', columns='Month', values='Precipitation')
Lutetian_prec_wide.columns = [month_abbreviations[col - 1] for col in Lutetian_prec_wide.columns]

# Reset the index to make 'Cell' a column again
Lutetian_SAT_wide.reset_index(inplace = True)
Lutetian_SST_wide.reset_index(inplace = True)
Lutetian_SSS_wide.reset_index(inplace = True)
Lutetian_prec_wide.reset_index(inplace = True)

# Merge the datasets of SAT, SST, SSS and prec, force suffixes to be added to the column names
# Merge in two steps to circumvent different numbers of cells due to differing spatial resolution in air and ocean models
Lutetian_models = pd.merge(
    # Merge SAT and prec data
    pd.merge(
        Lutetian_SAT_wide.rename(columns = {c: c+'_SAT' for c in Lutetian_SAT_wide.columns if c != 'Cell'}),
        Lutetian_prec_wide.rename(columns = {c: c+'_precip' for c in Lutetian_prec_wide.columns if c != 'Cell'}),
        on = 'Cell',
        how = 'outer'
    ),
    # Merge SST and SSS data
    pd.merge(
        Lutetian_SST_wide.rename(columns = {c: c+'_SST' for c in Lutetian_SST_wide.columns if c != 'Cell'}),
        Lutetian_SSS_wide.rename(columns = {c: c+'_SSS' for c in Lutetian_SSS_wide.columns if c != 'Cell'}),
        on = 'Cell',
        how = 'outer'
    ),
    on = 'Cell',
    how = 'outer'
)

# Display the combined dataset
Lutetian_models.head()

# Create list of month names
months = ['ja', 'fb', 'mr', 'ar', 'my', 'jn', 'jl', 'ag', 'sp', 'ot', 'nv', 'dc']

# Prior SST, SAT, SSS & precipitation estimates from climate models (mean)
mu_prior_SAT_monthly = np.array(Lutetian_models[[f"{month}_SAT" for month in months]].mean(axis=0, skipna=True))
mu_prior_SST_monthly = np.array(Lutetian_models[[f"{month}_SST" for month in months]].mean(axis=0, skipna=True))
mu_prior_SSS_monthly = np.array(Lutetian_models[[f"{month}_SSS" for month in months]].mean(axis=0, skipna=True))
mu_prior_precip_monthly = np.array(Lutetian_models[[f"{month}_precip" for month in months]].mean(axis=0, skipna=True))

# Covariance between months in prior SST, SAT, SSS, and precip estimates from climate models (covariance matrix)
cov_prior_SAT_monthly = np.cov(Lutetian_models[[f"{month}_SAT" for month in months]].dropna(), rowvar=False)
cov_prior_SST_monthly = np.cov(Lutetian_models[[f"{month}_SST" for month in months]].dropna(), rowvar=False)
cov_prior_SSS_monthly = np.cov(Lutetian_models[[f"{month}_SSS" for month in months]].dropna(), rowvar=False)
cov_prior_precip_monthly = np.cov(Lutetian_models[[f"{month}_precip" for month in months]].dropna(), rowvar=False)

# Store copy of original prior means to keep when later updating the prior
mu_prior_SAT_monthly_original, cov_prior_SAT_monthly_original = mu_prior_SAT_monthly.copy(), cov_prior_SAT_monthly.copy()
mu_prior_SST_monthly_original, cov_prior_SST_monthly_original = mu_prior_SST_monthly.copy(), cov_prior_SST_monthly.copy()
mu_prior_SSS_monthly_original, cov_prior_SSS_monthly_original = mu_prior_SSS_monthly.copy(), cov_prior_SSS_monthly.copy()
mu_prior_precip_monthly_original, cov_prior_precip_monthly_original = mu_prior_precip_monthly.copy(), cov_prior_precip_monthly.copy()

# Extract the standard deviations (uncertainty) from the covariance matrix
std_prior_SAT_monthly = np.sqrt(np.diag(cov_prior_SAT_monthly))
std_prior_SST_monthly = np.sqrt(np.diag(cov_prior_SST_monthly))
std_prior_SSS_monthly = np.sqrt(np.diag(cov_prior_SSS_monthly))
std_prior_precip_monthly = np.sqrt(np.diag(cov_prior_precip_monthly))

print("SAT Monthly Means:", mu_prior_SAT_monthly)
print("SAT Monthly Std Devs:", std_prior_SAT_monthly)
print("SST Monthly Means:", mu_prior_SST_monthly)
print("SST Monthly Std Devs:", std_prior_SST_monthly)
print("SSS Monthly Means:", mu_prior_SSS_monthly)
print("SSS Monthly Std Devs:", std_prior_SSS_monthly)
print("Precip Monthly Means:", mu_prior_precip_monthly)
print("Precip Monthly Std Devs:", std_prior_precip_monthly)

SAT Monthly Means: [16.27954224 17.01156006 18.66868896 21.54889648 27.12244995 32.0811853
 35.76927734 35.78623779 32.02878784 25.72896606 20.42561768 17.17369751]
SAT Monthly Std Devs: [3.46377896 3.10594374 2.78918732 2.54271169 2.46488387 2.73331005
 2.85535749 2.646315   2.52596201 3.00791784 3.58824735 3.69933077]
SST Monthly Means: [23.86139364 23.17810472 23.14772474 23.92908666 26.57514352 30.58667586
 34.10417351 35.48879139 34.18526063 31.27739675 28.24302759 25.51228172]
SST Monthly Std Devs: [2.83903042 2.98635231 2.95864974 2.74006009 2.42216397 2.22547687
 2.02150675 1.9040936  1.73710814 1.64930877 2.07319756 2.54273385]
SSS Monthly Means: [34.48952147 34.48138433 34.46366809 34.44571124 34.43899317 34.43806975
 34.41677989 34.40664036 34.40840345 34.436755   34.46977234 34.49277948]
SSS Monthly Std Devs: [3.01539682 2.98680818 2.97232063 2.97675101 3.00281989 3.03418516
 3.0716477  3.107458   3.13625919 3.14157102 3.10472103 3.05326323]
Precip Monthly Means: [0.20503649 0.20900211 0.21310989 0.19671564 0.14064345 0.16901619
 0.20951558 0.19801853 0.19421264 0.16398235 0.18276501 0.19433419]
Precip Monthly Std Devs: [0.06656403 0.06261988 0.05604318 0.06767304 0.07139255 0.09812866
 0.12533398 0.10071089 0.0756273  0.0488595  0.0686841  0.0680282 ]

# Set dimensions of data
n_models_monthly = len(Lutetian_models["Cell"])  # Find the total number of models

# Create a monthly scale for the x-axis
month_names = ['January', 'February', 'March', 'April', 'May', 'June', 'July', 'August', 'September', 'October', 'November', 'December']  # List full month names
months_scale = np.arange(len(months)) + 1  # Create monthly scale

# Create the figure and axes
fig, axes = plt.subplots(2, 1, figsize=(10, 12), sharex=True)

# Panel 1: Plot the prior distribution for SST and SAT
axes[0].plot(months_scale, mu_prior_SAT_monthly, label='Prior SAT Mean', marker='o', color='r')
axes[0].plot(months_scale, mu_prior_SST_monthly, label='Prior SST Mean', marker='o', color='b')

# Add 95% confidence intervals for SAT
axes[0].fill_between(
    months_scale,
    mu_prior_SAT_monthly - stats.t.ppf(1 - 0.025, n_models_monthly) * std_prior_SAT_monthly / np.sqrt(n_models_monthly),
    mu_prior_SAT_monthly + stats.t.ppf(1 - 0.025, n_models_monthly) * std_prior_SAT_monthly / np.sqrt(n_models_monthly),
    alpha=0.2, color='r', label='SAT 95% CI'
)

# Add 95% confidence intervals for SST
axes[0].fill_between(
    months_scale,
    mu_prior_SST_monthly - stats.t.ppf(1 - 0.025, n_models_monthly) * std_prior_SST_monthly / np.sqrt(n_models_monthly),
    mu_prior_SST_monthly + stats.t.ppf(1 - 0.025, n_models_monthly) * std_prior_SST_monthly / np.sqrt(n_models_monthly),
    alpha=0.2, color='b', label='SST 95% CI'
)

axes[0].set_title('Prior Mean and 95% Confidence Interval for Monthly SST & SAT Values')
axes[0].set_ylabel('Temperature (°C)')
axes[0].legend()
axes[0].grid(True)

# Panel 2: Plot the prior distribution for SSS and precipitation
axes[1].plot(months_scale, mu_prior_SSS_monthly, label='Prior SSS Mean', marker='o', color='g')
ax2 = axes[1].twinx()  # Create a secondary y-axis for precipitation
ax2.plot(months_scale, mu_prior_precip_monthly, label='Prior Precipitation Mean', marker='o', color='purple')

# Add 95% confidence intervals for SSS
axes[1].fill_between(
    months_scale,
    mu_prior_SSS_monthly - stats.t.ppf(1 - 0.025, n_models_monthly) * std_prior_SSS_monthly / np.sqrt(n_models_monthly),
    mu_prior_SSS_monthly + stats.t.ppf(1 - 0.025, n_models_monthly) * std_prior_SSS_monthly / np.sqrt(n_models_monthly),
    alpha=0.2, color='g', label='SSS 95% CI'
)

# Add 95% confidence intervals for precipitation
ax2.fill_between(
    months_scale,
    mu_prior_precip_monthly - stats.t.ppf(1 - 0.025, n_models_monthly) * std_prior_precip_monthly / np.sqrt(n_models_monthly),
    mu_prior_precip_monthly + stats.t.ppf(1 - 0.025, n_models_monthly) * std_prior_precip_monthly / np.sqrt(n_models_monthly),
    alpha=0.2, color='purple', label='Precipitation 95% CI'
)

axes[1].set_ylabel('SSS (psu)', color='g')
ax2.set_ylabel('Precipitation (mm/day)', color='purple')
axes[1].set_title('Prior Mean and 95% Confidence Interval for Monthly SSS & Precipitation Values')
axes[1].legend(loc='upper left')
ax2.legend(loc='upper right')
axes[1].grid(True)

# Update the x-axis with month names
plt.xticks(months_scale, month_names, rotation=45, ha="right")
plt.xlabel('Month')
plt.tight_layout()
plt.show()

# Apply T47()-function from the D47calib package to all SAT columns
# Identify the SAT columns
SAT_columns = [col for col in Lutetian_models.columns if col.endswith('_SAT')]
SST_columns = [col for col in Lutetian_models.columns if col.endswith('_SST')]

# Apply the conversion function to the SAT columns and add new columns for D47 and D47_SE
for col in SAT_columns:
    base_col_name = col.replace('_SAT', '') # Remove the '_SAT' suffix from the column name
    Lutetian_models[f'{base_col_name}_SAT_D47'], Lutetian_models[f'{base_col_name}_SAT_D47_SE'] = zip(*Lutetian_models[col].apply(
        lambda x: D47c.OGLS23.T47(T = x) if not pd.isna(x) else (np.nan, np.nan)
    )) # Use zip() to unpack the tuple returned by the apply() method and apply the T47()-function to each value in the column
for col in SST_columns:
    base_col_name = col.replace('_SST', '') # Remove the '_SST' suffix from the column name
    Lutetian_models[f'{base_col_name}_SST_D47'], Lutetian_models[f'{base_col_name}_SST_D47_SE'] = zip(*Lutetian_models[col].apply(
        lambda x: D47c.OGLS23.T47(T = x) if not pd.isna(x) else (np.nan, np.nan)
    )) # Use zip() to unpack the tuple returned by the apply() method and apply the T47()-function to each value in the column

# Display the combined data with D47 and D47_SE columns
D47_columns = [col for col in Lutetian_models.columns if col.endswith('_D47')]
D47_se_columns = [col for col in Lutetian_models.columns if '_D47_SE' in col]
print("D47 values for all model outcomes:\n", Lutetian_models[D47_columns].head())
print("Calibration standard errors for all model outcomes:\n", Lutetian_models[D47_se_columns].head())

D47 values for all model outcomes:
    ja_SAT_D47  fb_SAT_D47  mr_SAT_D47  ar_SAT_D47  my_SAT_D47  jn_SAT_D47  \
0    0.637865    0.635067    0.629115    0.619502    0.601307    0.584664   
1    0.603134    0.602008    0.598014    0.591188    0.578009    0.564439   
2    0.602866    0.601969    0.598808    0.592691    0.580329    0.567315   
3    0.601023    0.600939    0.598426    0.592938    0.581667    0.568903   
4    0.604475    0.603903    0.599502    0.591387    0.576345    0.561918   

   jl_SAT_D47  ag_SAT_D47  sp_SAT_D47  ot_SAT_D47  ...  mr_SST_D47  \
0    0.575942    0.575647    0.588334    0.607248  ...    0.593324   
1    0.554885    0.553952    0.563022    0.577000  ...    0.594240   
2    0.556444    0.554857    0.563225    0.577214  ...    0.595136   
3    0.557261    0.555379    0.563453    0.576950  ...    0.618591   
4    0.549188    0.549164    0.560136    0.577425  ...    0.618968   

   ar_SST_D47  my_SST_D47  jn_SST_D47  jl_SST_D47  ag_SST_D47  sp_SST_D47  \
0    0.591086    0.583574    0.572417    0.564044    0.559876    0.562770   
1    0.592472    0.585470    0.574402    0.566375    0.562104    0.564462   
2    0.593459    0.586555    0.575418    0.567460    0.563169    0.565401   
3    0.615362    0.605418    0.592590    0.582142    0.577727    0.579656   
4    0.615934    0.606080    0.593130    0.582374    0.577869    0.579768   

   ot_SST_D47  nv_SST_D47  dc_SST_D47  
0    0.570493    0.579090    0.586671  
1    0.571625    0.580172    0.587853  
2    0.572431    0.581026    0.589019  
3    0.586429    0.597502    0.608002  
4    0.586786    0.597919    0.608058  

[5 rows x 24 columns]
Calibration standard errors for all model outcomes:
    ja_SAT_D47_SE  fb_SAT_D47_SE  mr_SAT_D47_SE  ar_SAT_D47_SE  my_SAT_D47_SE  \
0       0.001252       0.001233       0.001196       0.001144       0.001073   
1       0.001079       0.001075       0.001065       0.001051       0.001041   
2       0.001078       0.001075       0.001067       0.001054       0.001041   
3       0.001073       0.001072       0.001066       0.001054       0.001042   
4       0.001083       0.001081       0.001069       0.001052       0.001041   

   jn_SAT_D47_SE  jl_SAT_D47_SE  ag_SAT_D47_SE  sp_SAT_D47_SE  ot_SAT_D47_SE  \
0       0.001044       0.001041       0.001041       0.001047       0.001092   
1       0.001051       0.001070       0.001072       0.001053       0.001041   
2       0.001047       0.001066       0.001070       0.001053       0.001041   
3       0.001046       0.001064       0.001069       0.001053       0.001041   
4       0.001055       0.001085       0.001085       0.001059       0.001041   

   ...  mr_SST_D47_SE  ar_SST_D47_SE  my_SST_D47_SE  jn_SST_D47_SE  \
0  ...       0.001055       0.001051       0.001043       0.001043   
1  ...       0.001057       0.001053       0.001044       0.001042   
2  ...       0.001058       0.001055       0.001045       0.001041   
3  ...       0.001140       0.001125       0.001086       0.001054   
4  ...       0.001141       0.001127       0.001088       0.001055   

   jl_SST_D47_SE  ag_SST_D47_SE  sp_SST_D47_SE  ot_SST_D47_SE  nv_SST_D47_SE  \
0       0.001052       0.001059       0.001054       0.001044       0.001041   
1       0.001049       0.001055       0.001051       0.001043       0.001041   
2       0.001047       0.001053       0.001050       0.001043       0.001042   
3       0.001042       0.001041       0.001041       0.001045       0.001064   
4       0.001042       0.001041       0.001041       0.001046       0.001065   

   dc_SST_D47_SE  
0       0.001045  
1       0.001047  
2       0.001048  
3       0.001095  
4       0.001095  

[5 rows x 24 columns]

# Apply the d18Ow-SSS function from Harwood et al. (2007) to all SSS columns
# Identify the SSS columns
SSS_columns = [col for col in Lutetian_models.columns if col.endswith('_SSS')]

# Apply the conversion function to the SSS columns and add new columns for d18Ow and d18Ow_SE
for col in SSS_columns:
    base_col_name = col.replace('_SSS', '')  # Remove the '_SSS' suffix from the column name
    Lutetian_models[f'{base_col_name}_SSS_d18Ow'] = Lutetian_models[col].apply(
        lambda x: -9.300 + 0.274 * x if not pd.isna(x) else np.nan  # Calculate d18Ow
    )

# Display the combined data with d18Ow and d18Ow_SE columns
d18Ow_columns = [col for col in Lutetian_models.columns if col.endswith('_d18Ow')]
print("d18Ow values for all model outcomes:\n", Lutetian_models[d18Ow_columns].head())

d18Ow values for all model outcomes:
    ja_SSS_d18Ow  fb_SSS_d18Ow  mr_SSS_d18Ow  ar_SSS_d18Ow  my_SSS_d18Ow  \
0      0.412163      0.417397      0.416426      0.420510      0.427636   
1      0.412551      0.419215      0.420105      0.424757      0.431978   
2      0.391470      0.397124      0.397364      0.401573      0.409009   
3     -1.716298     -1.696593     -1.699113     -1.718610     -1.746921   
4     -1.724499     -1.705170     -1.711377     -1.735129     -1.771551   

   jn_SSS_d18Ow  jl_SSS_d18Ow  ag_SSS_d18Ow  sp_SSS_d18Ow  ot_SSS_d18Ow  \
0      0.433800      0.432326      0.418661      0.386762      0.361828   
1      0.436819      0.435099      0.426873      0.402075      0.378333   
2      0.412444      0.409529      0.403192      0.382842      0.365661   
3     -1.771086     -1.798713     -1.820124     -1.836465     -1.833743   
4     -1.802794     -1.832529     -1.854027     -1.869227     -1.862037   

   nv_SSS_d18Ow  dc_SSS_d18Ow  
0      0.371852      0.395830  
1      0.379386      0.397893  
2      0.365822      0.380555  
3     -1.797647     -1.748480  
4     -1.814293     -1.760033

# Iterate over each model and calculate d18Oc values
for index, row in Lutetian_models.iterrows():
    # Iterate over each month
    for month in months:
        SST = row[f"{month}_SST"]
        d18Ow = row[f"{month}_SSS_d18Ow"]
        if not pd.isna(SST) and not pd.isna(d18Ow):
            d18Oc = (20.6 - SST) / 4.34 + (d18Ow - 0.27)
        else:
            d18Oc = np.nan
        # Add the calculated d18Oc value to the DataFrame
        Lutetian_models.loc[index, f"{month}_d18Oc"] = d18Oc

# Display the updated DataFrame
Lutetian_models.head()

# Set the weights of the data based on the standard errors
weights_monthly_SST_D47 = 1 / Lutetian_models[[f"{month}_SST_D47_SE" for month in months]] ** 2
weights_monthly_SAT_D47 = 1 / Lutetian_models[[f"{month}_SAT_D47_SE" for month in months]] ** 2

# Change the column suffixes from "_D47_SE" to "_D47" in weights_monthly_SST_D47 to match the headers of the D47 matrix later for multiplication
weights_monthly_SST_D47.columns = [col.replace('_SST_D47_SE', '_SST_D47') for col in weights_monthly_SST_D47.columns]
weights_monthly_SAT_D47.columns = [col.replace('_SAT_D47_SE', '_SAT_D47') for col in weights_monthly_SAT_D47.columns]

# Prior D47 estimates from climate models (weighted mean)
mu_prior_SST_D47_monthly = np.array((Lutetian_models[[f"{month}_SST_D47" for month in months]] * weights_monthly_SST_D47).sum(axis = 0, skipna = True) / weights_monthly_SST_D47.sum(axis = 0, skipna = True)) # Calculate weighted monthly mean D47 values and convert to numpy array
mu_prior_SAT_D47_monthly = np.array((Lutetian_models[[f"{month}_SAT_D47" for month in months]] * weights_monthly_SAT_D47).sum(axis = 0, skipna = True) / weights_monthly_SAT_D47.sum(axis = 0, skipna = True)) # Calculate weighted monthly mean D47 values and convert to numpy array

# Calculate simple (unweighted) mean for monthly d18Ow values
mu_prior_SSS_d18Ow_monthly = np.array(Lutetian_models[[f"{month}_SSS_d18Ow" for month in months]].mean(axis=0, skipna=True))
mu_prior_d18Oc_monthly = np.array(Lutetian_models[[f"{month}_d18Oc" for month in months]].mean(axis=0, skipna=True))

# Decompose variance within and between model outcomes
model_variances_SST = Lutetian_models[[f"{month}_SST_D47" for month in months]].var(axis = 0, ddof = 1)  # Compute variance across models
model_variances_SAT = Lutetian_models[[f"{month}_SAT_D47" for month in months]].var(axis = 0, ddof = 1)  # Compute variance across models
model_variances_d18Ow = Lutetian_models[[f"{month}_SSS_d18Ow" for month in months]].var(axis = 0, ddof = 1)  # Compute variance across models
model_variances_d18Oc = Lutetian_models[[f"{month}_d18Oc" for month in months]].var(axis = 0, ddof = 1)  # Compute variance across models
measurement_variances_SST = (Lutetian_models[[f"{month}_SST_D47_SE" for month in months]] ** 2).mean(axis = 0, skipna = True)  # Compute variance on measurements
measurement_variances_SAT = (Lutetian_models[[f"{month}_SAT_D47_SE" for month in months]] ** 2).mean(axis = 0, skipna = True)  # Compute variance on measurements

# Covariance between months in prior D47 estimates from climate models (weighted covariance matrix)
cov_raw_monthly_SST = np.cov(Lutetian_models[[f"{month}_SST_D47" for month in months]].dropna(), rowvar = False)  # Compute the covariance matrix for the raw data (without measurement uncertainty)
cov_raw_monthly_SAT = np.cov(Lutetian_models[[f"{month}_SAT_D47" for month in months]].dropna(), rowvar = False)  # Compute the covariance matrix for the raw data (without measurement uncertainty)
cov_raw_monthly_d18Ow = np.cov(Lutetian_models[[f"{month}_SSS_d18Ow" for month in months]].dropna(), rowvar = False)  # Compute the covariance matrix for the raw data (without measurement uncertainty)
cov_raw_monthly_d18Oc = np.cov(Lutetian_models[[f"{month}_d18Oc" for month in months]].dropna(), rowvar = False)  # Compute the covariance matrix for the raw data (without measurement uncertainty)
cov_prior_SST_D47_monthly = cov_raw_monthly_SST.copy() # Copy covariance matrix to add uncertainty coming from the measurements
cov_prior_SAT_D47_monthly = cov_raw_monthly_SAT.copy() # Copy covariance matrix to add uncertainty coming from the measurements
np.fill_diagonal(cov_prior_SST_D47_monthly, np.diagonal(cov_raw_monthly_SST) + measurement_variances_SST)  # Add diagonal terms for measurement uncertainties (which have no covariance between models)
np.fill_diagonal(cov_prior_SAT_D47_monthly, np.diagonal(cov_raw_monthly_SAT) + measurement_variances_SAT)  # Add diagonal terms for measurement uncertainties (which have no covariance between models)

# Store copy of original prior means to keep when later updating the prior
mu_prior_SST_D47_monthly_original, cov_prior_SST_D47_monthly_original = mu_prior_SST_D47_monthly.copy(), cov_prior_SST_D47_monthly.copy()
mu_prior_SAT_D47_monthly_original, cov_prior_SAT_D47_monthly_original = mu_prior_SAT_D47_monthly.copy(), cov_prior_SAT_D47_monthly.copy()
mu_prior_SSS_d18Ow_monthly_original, cov_prior_SSS_d18Ow_monthly_original = mu_prior_SSS_d18Ow_monthly.copy(), cov_raw_monthly_d18Ow.copy()
mu_prior_d18Oc_monthly_original, cov_prior_d18Oc_monthly_original = mu_prior_d18Oc_monthly.copy(), cov_raw_monthly_d18Oc.copy()

# Extract the standard deviations (uncertainty) from the covariance matrix
std_prior_SST_D47_monthly = np.sqrt(np.diag(cov_prior_SST_D47_monthly))
std_prior_SAT_D47_monthly = np.sqrt(np.diag(cov_prior_SAT_D47_monthly))
std_prior_SSS_d18Ow_monthly = np.sqrt(np.diag(cov_raw_monthly_d18Ow))
std_prior_d18Oc_monthly = np.sqrt(np.diag(cov_raw_monthly_d18Oc))

# Print the results
print("Prior D47 estimates from SST in climate models (weighted mean):")
print(mu_prior_SST_D47_monthly)
print("Prior D47 estimates from SST in climate models (weighted covariance matrix):")
print(std_prior_SST_D47_monthly)
print("Prior D47 estimates from SAT in climate models (weighted mean):")
print(mu_prior_SAT_D47_monthly)
print("Prior D47 estimates from SAT in climate models (weighted covariance matrix):")
print(std_prior_SAT_D47_monthly)
print("Prior d18Ow estimates from SSS in climate models (weighted mean):")
print(mu_prior_SSS_d18Ow_monthly)
print("Prior d18Ow estimates from SSS in climate models (weighted covariance matrix):")
print(std_prior_SSS_d18Ow_monthly)
print("Prior d18Oc estimates from SST and d18Ow in climate models (weighted mean):")
print(mu_prior_d18Oc_monthly)
print("Prior d18Oc estimates from SST and d18Ow in climate models (weighted covariance matrix):")
print(std_prior_d18Oc_monthly)

Prior D47 estimates from SST in climate models (weighted mean):
[0.59656789 0.59858139 0.59867924 0.59638204 0.58858685 0.57705811
 0.56727922 0.56351195 0.5670409  0.57510508 0.58376364 0.59171077]
Prior D47 estimates from SST in climate models (weighted covariance matrix):
[0.00890411 0.00944467 0.00936157 0.00859554 0.00739686 0.00653869
 0.00575527 0.00534817 0.00493985 0.00482531 0.00621776 0.00782822]
Prior D47 estimates from SAT in climate models (weighted mean):
[0.62006985 0.61787124 0.61269573 0.60374682 0.58703137 0.57294061
 0.56290407 0.56283341 0.57307851 0.59111554 0.60696862 0.61709809]
Prior D47 estimates from SAT in climate models (weighted covariance matrix):
[0.01142741 0.01018869 0.0090148  0.00801557 0.00738831 0.00779214
 0.00785649 0.00725533 0.00718243 0.00905268 0.01135674 0.01208698]
Prior d18Ow estimates from SSS in climate models (weighted mean):
[0.15012888 0.14789931 0.14304506 0.13812488 0.13628413 0.13603111
 0.13019769 0.12741946 0.12790254 0.13567087 0.14471762 0.15102158]
Prior d18Ow estimates from SSS in climate models (weighted covariance matrix):
[0.82621873 0.81838544 0.81441585 0.81562978 0.82277265 0.83136673
 0.84163147 0.85144349 0.85933502 0.86079046 0.85069356 0.83659413]
Prior d18Oc estimates from SST and d18Ow in climate models (weighted mean):
[-0.87134431 -0.71613404 -0.7139883  -0.89894578 -1.51047705 -2.43504627
 -3.25136303 -3.57317763 -3.27234184 -2.5945588  -1.88634864 -1.25084057]
Prior d18Oc estimates from SST and d18Ow in climate models (weighted covariance matrix):
[0.43074162 0.4079074  0.39792762 0.39889866 0.41067086 0.41700845
 0.43048772 0.48872905 0.56171093 0.58936283 0.51749387 0.46825537]

# Plot monthly prior distribution
fig, axes = plt.subplots(2, 2, figsize=(15, 12))  # Adjust the figure to have 2x2 grid

# Plot the prior distribution for SST
axes[0, 0].plot(months_scale, mu_prior_SST_D47_monthly, label='Prior SST Mean', color='b', marker='o')
axes[0, 0].fill_between(months_scale,
                        mu_prior_SST_D47_monthly - stats.t.ppf(1 - 0.025, n_models_monthly) * std_prior_SST_D47_monthly / np.sqrt(n_models_monthly),
                        mu_prior_SST_D47_monthly + stats.t.ppf(1 - 0.025, n_models_monthly) * std_prior_SST_D47_monthly / np.sqrt(n_models_monthly),
                        color='b', alpha=0.2, label='95% Confidence Interval')
axes[0, 0].set_xticks(months_scale)
axes[0, 0].set_xticklabels(month_names, rotation=45, ha="right")
axes[0, 0].set_title('Prior Mean and 95% Confidence Interval for Monthly SST D47 values')
axes[0, 0].set_xlabel('Month')
axes[0, 0].set_ylabel('D47 value')
axes[0, 0].legend()
axes[0, 0].grid(True)

# Plot the prior distribution for SAT
axes[0, 1].plot(months_scale, mu_prior_SAT_D47_monthly, label='Prior SAT Mean', color='r', marker='o')
axes[0, 1].fill_between(months_scale,
                        mu_prior_SAT_D47_monthly - stats.t.ppf(1 - 0.025, n_models_monthly) * std_prior_SAT_D47_monthly / np.sqrt(n_models_monthly),
                        mu_prior_SAT_D47_monthly + stats.t.ppf(1 - 0.025, n_models_monthly) * std_prior_SAT_D47_monthly / np.sqrt(n_models_monthly),
                        color='r', alpha=0.2, label='95% Confidence Interval')
axes[0, 1].set_xticks(months_scale)
axes[0, 1].set_xticklabels(month_names, rotation=45, ha="right")
axes[0, 1].set_title('Prior Mean and 95% Confidence Interval for Monthly SAT D47 values')
axes[0, 1].set_xlabel('Month')
axes[0, 1].set_ylabel('D47 value')
axes[0, 1].legend()
axes[0, 1].grid(True)

# Plot the prior distribution for d18Oc
axes[1, 0].plot(months_scale, mu_prior_d18Oc_monthly, label='Prior d18Oc Mean', color='purple', marker='o')
axes[1, 0].fill_between(months_scale,
                        mu_prior_d18Oc_monthly - stats.t.ppf(1 - 0.025, n_models_monthly) * std_prior_d18Oc_monthly / np.sqrt(n_models_monthly),
                        mu_prior_d18Oc_monthly + stats.t.ppf(1 - 0.025, n_models_monthly) * std_prior_d18Oc_monthly / np.sqrt(n_models_monthly),
                        color='purple', alpha=0.2, label='95% Confidence Interval')
axes[1, 0].set_xticks(months_scale)
axes[1, 0].set_xticklabels(month_names, rotation=45, ha="right")
axes[1, 0].set_title('Prior Mean and 95% Confidence Interval for Monthly d18Oc values')
axes[1, 0].set_xlabel('Month')
axes[1, 0].set_ylabel('d18Oc value')
axes[1, 0].legend()
axes[1, 0].grid(True)

# Plot the prior distribution for precipitation
axes[1, 1].plot(months_scale, mu_prior_precip_monthly, label='Prior Precipitation Mean', color='teal', marker='o')
axes[1, 1].fill_between(months_scale,
                        mu_prior_precip_monthly - stats.t.ppf(1 - 0.025, n_models_monthly) * std_prior_precip_monthly / np.sqrt(n_models_monthly),
                        mu_prior_precip_monthly + stats.t.ppf(1 - 0.025, n_models_monthly) * std_prior_precip_monthly / np.sqrt(n_models_monthly),
                        color='teal', alpha=0.2, label='95% Confidence Interval')
axes[1, 1].set_xticks(months_scale)
axes[1, 1].set_xticklabels(month_names, rotation=45, ha="right")
axes[1, 1].set_title('Prior Mean and 95% Confidence Interval for Monthly Precipitation values')
axes[1, 1].set_xlabel('Month')
axes[1, 1].set_ylabel('Precipitation (mm/day)')
axes[1, 1].legend()
axes[1, 1].grid(True)

# Update the layout and show the plot
plt.tight_layout()
plt.show()

# Define column names for SAT, SST, d18Oc, and precipitation
SAT_D47_columns_monthly = [f"{month}_SAT_D47" for month in months]
SST_D47_columns_monthly = [f"{month}_SST_D47" for month in months]
d18Oc_columns_monthly = [f"{month}_d18Oc" for month in months]
precip_columns_monthly = [f"{month}_precip" for month in months]

# Extract the relevant columns for SAT, SST D47, d18Oc, and precipitation
SAT_D47_columns_monthly = [f"{month}_SAT_D47" for month in months]
SST_D47_columns_monthly = [f"{month}_SST_D47" for month in months]
d18Oc_columns_monthly = [f"{month}_d18Oc" for month in months]
precip_columns_monthly = [f"{month}_precip" for month in months]

# Combine the relevant columns into a single dataframe
combined_data_monthly = Lutetian_models[SAT_D47_columns_monthly + SST_D47_columns_monthly + d18Oc_columns_monthly + precip_columns_monthly]

# Calculate the covariance matrix for the combined data
cov_combined_monthly = np.cov(combined_data_monthly.dropna(), rowvar=False)

# Plot the heatmap of the raw combined covariance matrix
plt.figure(figsize=(12, 10))
sns.heatmap(
    cov_combined_monthly,  # Use the raw covariance matrix
    annot=False,
    fmt=".2f",
    cmap="coolwarm",
    center=0,
    xticklabels=SAT_D47_columns_monthly + SST_D47_columns_monthly + d18Oc_columns_monthly + precip_columns_monthly,
    yticklabels=SAT_D47_columns_monthly + SST_D47_columns_monthly + d18Oc_columns_monthly + precip_columns_monthly
)

# Add titles to the axes per parameter
plt.axvline(x=len(SAT_D47_columns_monthly), color='black', linestyle='--', linewidth=1)
plt.axvline(x=len(SAT_D47_columns_monthly) + len(SST_D47_columns_monthly), color='black', linestyle='--', linewidth=1)
plt.axvline(x=len(SAT_D47_columns_monthly) + len(SST_D47_columns_monthly) + len(d18Oc_columns_monthly), color='black', linestyle='--', linewidth=1)

plt.axhline(y=len(SAT_D47_columns_monthly), color='black', linestyle='--', linewidth=1)
plt.axhline(y=len(SAT_D47_columns_monthly) + len(SST_D47_columns_monthly), color='black', linestyle='--', linewidth=1)
plt.axhline(y=len(SAT_D47_columns_monthly) + len(SST_D47_columns_monthly) + len(d18Oc_columns_monthly), color='black', linestyle='--', linewidth=1)

# Add parameter labels
plt.text(len(SAT_D47_columns_monthly) / 2, -2, 'D47 value from SAT', ha='center', va='center', fontsize=10)
plt.text(len(SAT_D47_columns_monthly) + len(SST_D47_columns_monthly) / 2, -2, 'D47 value from SST', ha='center', va='center', fontsize=10)
plt.text(len(SAT_D47_columns_monthly) + len(SST_D47_columns_monthly) + len(d18Oc_columns_monthly) / 2, -2, 'd18Oc', ha='center', va='center', fontsize=10)
plt.text(len(SAT_D47_columns_monthly) + len(SST_D47_columns_monthly) + len(d18Oc_columns_monthly) + len(precip_columns_monthly) / 2, -2, 'Precipitation', ha='center', va='center', fontsize=10)

plt.text(-2, len(SAT_D47_columns_monthly) / 2, 'D47 value from SAT', ha='center', va='center', rotation=90, fontsize=10)
plt.text(-2, len(SAT_D47_columns_monthly) + len(SST_D47_columns_monthly) / 2, 'D47 value from SST', ha='center', va='center', rotation=90, fontsize=10)
plt.text(-2, len(SAT_D47_columns_monthly) + len(SST_D47_columns_monthly) + len(d18Oc_columns_monthly) / 2, 'd18Oc', ha='center', va='center', rotation=90, fontsize=10)
plt.text(-2, len(SAT_D47_columns_monthly) + len(SST_D47_columns_monthly) + len(d18Oc_columns_monthly) + len(precip_columns_monthly) / 2, 'Precipitation', ha='center', va='center', rotation=90, fontsize=10)

plt.title("Raw Combined Covariance Matrix")
plt.show()

# Normalize each submatrix independently for better visualization
def normalize_matrix(matrix):
    min_val = np.min(matrix)
    max_val = np.max(matrix)
    return (matrix - min_val) / (max_val - min_val)

# Extract the relevant columns for SAT, SST D47, d18Oc, and precipitation
SAT_D47_columns_monthly = [f"{month}_SAT_D47" for month in months]
SST_D47_columns_monthly = [f"{month}_SST_D47" for month in months]
d18Oc_columns_monthly = [f"{month}_d18Oc" for month in months]
precip_columns_monthly = [f"{month}_precip" for month in months]

# Combine the relevant columns into a single dataframe
combined_data_monthly = Lutetian_models[SAT_D47_columns_monthly + SST_D47_columns_monthly + d18Oc_columns_monthly + precip_columns_monthly]

# Calculate the covariance matrix for the combined data
cov_combined_monthly = np.cov(combined_data_monthly.dropna(), rowvar=False)

# Extract the covariance matrices for SAT D47, SST D47, d18Oc, and precipitation
cov_SAT_D47_monthly = cov_combined_monthly[:len(months), :len(months)]
cov_SST_D47_monthly = cov_combined_monthly[len(months):2*len(months), len(months):2*len(months)]
cov_d18Oc_monthly = cov_combined_monthly[2*len(months):3*len(months), 2*len(months):3*len(months)]
cov_precip_monthly = cov_combined_monthly[3*len(months):, 3*len(months):]

# Extract the cross-covariance matrices
cross_cov_SAT_SST_D47_monthly = cov_combined_monthly[:len(months), len(months):2*len(months)]
cross_cov_SAT_d18Oc_monthly = cov_combined_monthly[:len(months), 2*len(months):3*len(months)]
cross_cov_SAT_precip_monthly = cov_combined_monthly[:len(months), 3*len(months):]
cross_cov_SST_d18Oc_monthly = cov_combined_monthly[len(months):2*len(months), 2*len(months):3*len(months)]
cross_cov_SST_precip_monthly = cov_combined_monthly[len(months):2*len(months), 3*len(months):]
cross_cov_d18Oc_precip_monthly = cov_combined_monthly[2*len(months):3*len(months), 3*len(months):]

# Normalize each submatrix
normalized_cov_SAT_D47_monthly = normalize_matrix(cov_SAT_D47_monthly)
normalized_cov_SST_D47_monthly = normalize_matrix(cov_SST_D47_monthly)
normalized_cov_d18Oc_monthly = normalize_matrix(cov_d18Oc_monthly)
normalized_cov_precip_monthly = normalize_matrix(cov_precip_monthly)

# Normalize each cross-covariance matrix
normalized_cross_cov_SAT_SST_D47_monthly = normalize_matrix(cross_cov_SAT_SST_D47_monthly)
normalized_cross_cov_SAT_d18Oc_monthly = normalize_matrix(cross_cov_SAT_d18Oc_monthly)
normalized_cross_cov_SAT_precip_monthly = normalize_matrix(cross_cov_SAT_precip_monthly)
normalized_cross_cov_SST_d18Oc_monthly = normalize_matrix(cross_cov_SST_d18Oc_monthly)
normalized_cross_cov_SST_precip_monthly = normalize_matrix(cross_cov_SST_precip_monthly)
normalized_cross_cov_d18Oc_precip_monthly = normalize_matrix(cross_cov_d18Oc_precip_monthly)

# Combine the normalized submatrices into a single normalized covariance matrix
normalized_cov_combined_monthly = np.block([
    [normalized_cov_SAT_D47_monthly, normalized_cross_cov_SAT_SST_D47_monthly, normalized_cross_cov_SAT_d18Oc_monthly, normalized_cross_cov_SAT_precip_monthly],
    [normalized_cross_cov_SAT_SST_D47_monthly.T, normalized_cov_SST_D47_monthly, normalized_cross_cov_SST_d18Oc_monthly, normalized_cross_cov_SST_precip_monthly],
    [normalized_cross_cov_SAT_d18Oc_monthly.T, normalized_cross_cov_SST_d18Oc_monthly.T, normalized_cov_d18Oc_monthly, normalized_cross_cov_d18Oc_precip_monthly],
    [normalized_cross_cov_SAT_precip_monthly.T, normalized_cross_cov_SST_precip_monthly.T, normalized_cross_cov_d18Oc_precip_monthly.T, normalized_cov_precip_monthly]
])

# Plot the heatmap of the normalized combined covariance matrix
plt.figure(figsize=(12, 10))
sns.heatmap(
    normalized_cov_combined_monthly,
    annot=False,
    fmt=".2f",
    cmap="coolwarm",
    center=0,
    xticklabels=SAT_D47_columns_monthly + SST_D47_columns_monthly + d18Oc_columns_monthly + precip_columns_monthly,
    yticklabels=SAT_D47_columns_monthly + SST_D47_columns_monthly + d18Oc_columns_monthly + precip_columns_monthly
)

# Add titles to the axes per parameter
plt.axvline(x=len(SAT_D47_columns_monthly), color='black', linestyle='--', linewidth=1)
plt.axvline(x=len(SAT_D47_columns_monthly) + len(SST_D47_columns_monthly), color='black', linestyle='--', linewidth=1)
plt.axvline(x=len(SAT_D47_columns_monthly) + len(SST_D47_columns_monthly) + len(d18Oc_columns_monthly), color='black', linestyle='--', linewidth=1)

plt.axhline(y=len(SAT_D47_columns_monthly), color='black', linestyle='--', linewidth=1)
plt.axhline(y=len(SAT_D47_columns_monthly) + len(SST_D47_columns_monthly), color='black', linestyle='--', linewidth=1)
plt.axhline(y=len(SAT_D47_columns_monthly) + len(SST_D47_columns_monthly) + len(d18Oc_columns_monthly), color='black', linestyle='--', linewidth=1)

# Add parameter labels
plt.text(len(SAT_D47_columns_monthly) / 2, -2, 'D47 value from SAT', ha='center', va='center', fontsize=10)
plt.text(len(SAT_D47_columns_monthly) + len(SST_D47_columns_monthly) / 2, -2, 'D47 value from SST', ha='center', va='center', fontsize=10)
plt.text(len(SAT_D47_columns_monthly) + len(SST_D47_columns_monthly) + len(d18Oc_columns_monthly) / 2, -2, 'd18Oc', ha='center', va='center', fontsize=10)
plt.text(len(SAT_D47_columns_monthly) + len(SST_D47_columns_monthly) + len(d18Oc_columns_monthly) + len(precip_columns_monthly) / 2, -2, 'Precipitation', ha='center', va='center', fontsize=10)

plt.text(-7, len(SAT_D47_columns_monthly) / 2, 'D47 value from SAT', ha='center', va='center', rotation=90, fontsize=10)
plt.text(-7, len(SAT_D47_columns_monthly) + len(SST_D47_columns_monthly) / 2, 'D47 value from SST', ha='center', va='center', rotation=90, fontsize=10)
plt.text(-7, len(SAT_D47_columns_monthly) + len(SST_D47_columns_monthly) + len(d18Oc_columns_monthly) / 2, 'd18Oc', ha='center', va='center', rotation=90, fontsize=10)
plt.text(-7, len(SAT_D47_columns_monthly) + len(SST_D47_columns_monthly) + len(d18Oc_columns_monthly) + len(precip_columns_monthly) / 2, 'Precipitation', ha='center', va='center', rotation=90, fontsize=10)

plt.title("Normalized Combined Covariance Matrix")
plt.show()

# Combine the prior means of D47 and SAT into a single state vector
mu_prior_monthly_combined = np.concatenate((mu_prior_SST_D47_monthly, mu_prior_SAT_D47_monthly, mu_prior_d18Oc_monthly, mu_prior_precip_monthly))

# Combine the covariance matrices of D47 values of SST and SAT, including the cross-covariance
cov_prior_monthly_combined = cov_combined_monthly.copy()

# Define the seasons
seasons = {
    "winter": ["dc", "ja", "fb"],
    "spring": ["mr", "ar", "my"],
    "summer": ["jn", "jl", "ag"],
    "autumn": ["sp", "ot", "nv"],
}

# Stack monthly columns to create seasonal dataframes
# Initialize dictionaries to store seasonal data
Lutetian_models_seasonal_dict = {}

# Identify the columns to process (all except the modelname column)
columns_to_process = [col for col in Lutetian_models.columns if any(suffix in col for suffix in [
    '_SST', '_SAT', '_SST_D47', '_SST_D47_SE', '_SAT_D47', '_SAT_D47_SE',
    '_SSS', '_d18Oc', '_precip'
])]

# Process each season
for season, months in seasons.items():  # Iterate over the seasons and corresponding months
    for col in columns_to_process:  # Iterate over the columns to process
        base_col_name = col.split('_')[0]  # Extract the base column name
        suffix = '_'.join(col.split('_')[1:])  # Extract the suffix
        if base_col_name in months:  # Check if the column corresponds to the current season
            season_col_name = f"{season}_{suffix}"  # Create the new column name
            if season_col_name not in Lutetian_models_seasonal_dict:  # Check if the new column name already exists in the seasonal data
                Lutetian_models_seasonal_dict[season_col_name] = []  # If not, initialize a new column in the seasonal data means
            Lutetian_models_seasonal_dict[season_col_name].append(Lutetian_models[col])

# Combine the seasonal data into a single dataframe
Lutetian_models_seasonal = pd.DataFrame()
for season_col_name, data in Lutetian_models_seasonal_dict.items():
    # Concatenate the data for each season and reshape it properly
    concatenated_data = pd.concat(data, axis=0).reset_index(drop=True)
    Lutetian_models_seasonal[season_col_name] = concatenated_data

# Add model names
Lutetian_models_seasonal["Cell"] = np.tile(Lutetian_models["Cell"].values, 3)  # Repeat the model names for each season

# Display the new seasonal DataFrame
D47_columns_seasonal = [col for col in Lutetian_models_seasonal.columns if col.endswith('_D47')]
D47_se_columns_seasonal = [col for col in Lutetian_models_seasonal.columns if '_D47_SE' in col]
SSS_columns_seasonal = [col for col in Lutetian_models_seasonal.columns if col.endswith('_SSS')]
d18Oc_columns_seasonal = [col for col in Lutetian_models_seasonal.columns if col.endswith('_d18Oc')]
precip_columns_seasonal = [col for col in Lutetian_models_seasonal.columns if col.endswith('_precip')]

print("Seasonal D47 values for all SST model outcomes:\n", Lutetian_models_seasonal[D47_columns_seasonal].head())
print("Calibration standard errors for all SST model outcomes:\n", Lutetian_models_seasonal[D47_se_columns_seasonal].head())
print("Seasonal SSS values for all model outcomes:\n", Lutetian_models_seasonal[SSS_columns_seasonal].head())
print("Seasonal d18Oc values for all model outcomes:\n", Lutetian_models_seasonal[d18Oc_columns_seasonal].head())
print("Seasonal precipitation values for all model outcomes:\n", Lutetian_models_seasonal[precip_columns_seasonal].head())

Seasonal D47 values for all SST model outcomes:
    winter_SAT_D47  winter_SST_D47  spring_SAT_D47  spring_SST_D47  \
0        0.637865        0.591435        0.629115        0.593324   
1        0.603134        0.592468        0.598014        0.594240   
2        0.602866        0.593614        0.598808        0.595136   
3        0.601023        0.614886        0.598426        0.618591   
4        0.604475        0.614867        0.599502        0.618968   

   summer_SAT_D47  summer_SST_D47  autumn_SAT_D47  autumn_SST_D47  
0        0.584664        0.572417        0.588334        0.562770  
1        0.564439        0.574402        0.563022        0.564462  
2        0.567315        0.575418        0.563225        0.565401  
3        0.568903        0.592590        0.563453        0.579656  
4        0.561918        0.593130        0.560136        0.579768  
Calibration standard errors for all SST model outcomes:
    winter_SAT_D47_SE  winter_SST_D47_SE  spring_SAT_D47_SE  spring_SST_D47_SE  \
0           0.001252           0.001052           0.001196           0.001055   
1           0.001079           0.001053           0.001065           0.001057   
2           0.001078           0.001055           0.001067           0.001058   
3           0.001073           0.001123           0.001066           0.001140   
4           0.001083           0.001122           0.001069           0.001141   

   summer_SAT_D47_SE  summer_SST_D47_SE  autumn_SAT_D47_SE  autumn_SST_D47_SE  
0           0.001044           0.001043           0.001047           0.001054  
1           0.001051           0.001042           0.001053           0.001051  
2           0.001047           0.001041           0.001053           0.001050  
3           0.001046           0.001054           0.001053           0.001041  
4           0.001055           0.001055           0.001059           0.001041  
Seasonal SSS values for all model outcomes:
    winter_SSS  spring_SSS  summer_SSS  autumn_SSS
0   35.445849   35.461410   35.524816   35.353144
1   35.447266   35.474837   35.535836   35.409034
2   35.370330   35.391840   35.446876   35.338840
3   27.677744   27.740462   27.477787   27.239179
4   27.647815   27.695703   27.362065   27.119611
Seasonal d18Oc values for all model outcomes:
    winter_d18Oc  spring_d18Oc  summer_d18Oc  autumn_d18Oc
0     -1.022149     -0.873077     -2.513306     -3.367792
1     -0.942424     -0.799470     -2.347593     -3.208780
2     -0.875875     -0.754007     -2.289134     -3.148663
3     -1.415697     -1.136514     -3.116770     -4.195826
4     -1.425293     -1.122320     -3.107112     -4.219610
Seasonal precipitation values for all model outcomes:
    winter_precip  spring_precip  summer_precip  autumn_precip
0       0.352807       0.288893       0.275469       0.386743
1       0.114865       0.127827       0.139227       0.331272
2       0.225590       0.182669       0.130147       0.214673
3       0.224708       0.158838       0.030059       0.112521
4       0.214614       0.162401       0.051653       0.118162

# Prior estimates from climate models (mean)
mu_prior_SAT_seasonal = np.array(Lutetian_models_seasonal[[f"{season}_SAT" for season in seasons]].mean(axis=0, skipna=True))
mu_prior_SST_seasonal = np.array(Lutetian_models_seasonal[[f"{season}_SST" for season in seasons]].mean(axis=0, skipna=True))
mu_prior_SSS_seasonal = np.array(Lutetian_models_seasonal[[f"{season}_SSS" for season in seasons]].mean(axis=0, skipna=True))
mu_prior_precip_seasonal = np.array(Lutetian_models_seasonal[[f"{season}_precip" for season in seasons]].mean(axis=0, skipna=True))

# Covariance between seasons in prior estimates from climate models (covariance matrix)
cov_prior_SAT_seasonal = np.cov(Lutetian_models_seasonal[[f"{season}_SAT" for season in seasons]].dropna(), rowvar=False)
cov_prior_SST_seasonal = np.cov(Lutetian_models_seasonal[[f"{season}_SST" for season in seasons]].dropna(), rowvar=False)
cov_prior_SSS_seasonal = np.cov(Lutetian_models_seasonal[[f"{season}_SSS" for season in seasons]].dropna(), rowvar=False)
cov_prior_precip_seasonal = np.cov(Lutetian_models_seasonal[[f"{season}_precip" for season in seasons]].dropna(), rowvar=False)

# Store copy of original prior means to keep when later updating the prior
mu_prior_SAT_seasonal_original, cov_prior_SAT_seasonal_original = mu_prior_SAT_seasonal.copy(), cov_prior_SAT_seasonal.copy()
mu_prior_SST_seasonal_original, cov_prior_SST_seasonal_original = mu_prior_SST_seasonal.copy(), cov_prior_SST_seasonal.copy()
mu_prior_SSS_seasonal_original, cov_prior_SSS_seasonal_original = mu_prior_SSS_seasonal.copy(), cov_prior_SSS_seasonal.copy()
mu_prior_precip_seasonal_original, cov_prior_precip_seasonal_original = mu_prior_precip_seasonal.copy(), cov_prior_precip_seasonal.copy()

# Extract the standard deviations (uncertainty) from the covariance matrix
std_prior_SAT_seasonal = np.sqrt(np.diag(cov_prior_SAT_seasonal))
std_prior_SST_seasonal = np.sqrt(np.diag(cov_prior_SST_seasonal))
std_prior_SSS_seasonal = np.sqrt(np.diag(cov_prior_SSS_seasonal))
std_prior_precip_seasonal = np.sqrt(np.diag(cov_prior_precip_seasonal))

# Print the results
print("SAT Seasonal Means:", mu_prior_SAT_seasonal)
print("SAT Seasonal Std Devs:", std_prior_SAT_seasonal)
print("SST Seasonal Means:", mu_prior_SST_seasonal)
print("SST Seasonal Std Devs:", std_prior_SST_seasonal)
print("SSS Seasonal Means:", mu_prior_SSS_seasonal)
print("SSS Seasonal Std Devs:", std_prior_SSS_seasonal)
print("Precipitation Seasonal Means:", mu_prior_precip_seasonal)
print("Precipitation Seasonal Std Devs:", std_prior_precip_seasonal)

SAT Seasonal Means: [16.82159993 22.44667847 34.54556681 26.06112386]
SAT Seasonal Std Devs: [3.40758273 4.36702297 3.22740902 5.65487419]
SST Seasonal Means: [24.18392669 24.55065164 33.39321359 31.23522833]
SST Seasonal Std Devs: [2.95196389 3.07816593 2.90897972 3.03774427]
SSS Seasonal Means: [34.48789509 34.4494575  34.42049667 34.43831026]
SSS Seasonal Std Devs: [3.00611648 2.97165638 3.0585534  3.11471013]
Precipitation Seasonal Means: [0.20279093 0.18348966 0.19218344 0.18032   ]
Precipitation Seasonal Std Devs: [0.06518114 0.07164279 0.10863477 0.06570094]

# Define the seasons, number of models, and scale for the x-axis
seasons = ["winter", "spring", "summer", "autumn"]
n_models_seasonal = len(Lutetian_models["Cell"])  # Find the total number of models
seasons_scale = np.arange(len(seasons)) + 1  # Create seasonal scale

# Create a 1x2 plotting grid
fig, axes = plt.subplots(2, 1, figsize=(10, 10))

# Panel 1: Plot the prior distribution for SST and SAT
axes[0].plot(seasons_scale, mu_prior_SST_seasonal, label='Prior SST Mean', marker='o', color='b')
axes[0].fill_between(
    seasons_scale,
    mu_prior_SST_seasonal - stats.t.ppf(1 - 0.025, n_models_seasonal) * std_prior_SST_seasonal / np.sqrt(n_models_seasonal),
    mu_prior_SST_seasonal + stats.t.ppf(1 - 0.025, n_models_seasonal) * std_prior_SST_seasonal / np.sqrt(n_models_seasonal),
    alpha=0.2, color='b', label='SST 95% CI'
)
axes[0].plot(seasons_scale, mu_prior_SAT_seasonal, label='Prior SAT Mean', marker='o', color='r')
axes[0].fill_between(
    seasons_scale,
    mu_prior_SAT_seasonal - stats.t.ppf(1 - 0.025, n_models_seasonal) * std_prior_SAT_seasonal / np.sqrt(n_models_seasonal),
    mu_prior_SAT_seasonal + stats.t.ppf(1 - 0.025, n_models_seasonal) * std_prior_SAT_seasonal / np.sqrt(n_models_seasonal),
    alpha=0.2, color='r', label='SAT 95% CI'
)
axes[0].set_title('Prior Mean and 95% Confidence Interval for Seasonal SST & SAT')
axes[0].set_xlabel('Season')
axes[0].set_ylabel('Temperature (°C)')
axes[0].set_xticks(seasons_scale)
axes[0].set_xticklabels(seasons)
axes[0].legend()
axes[0].grid(True)

# Panel 2: Plot the prior distribution for SSS and precipitation
axes[1].plot(seasons_scale, mu_prior_SSS_seasonal, label='Prior SSS Mean', marker='o', color='g')
ax2 = axes[1].twinx()  # Create a secondary y-axis for precipitation
ax2.plot(seasons_scale, mu_prior_precip_seasonal, label='Prior Precipitation Mean', marker='o', color='purple')

# Add 95% confidence intervals for SSS
axes[1].fill_between(
    seasons_scale,
    mu_prior_SSS_seasonal - stats.t.ppf(1 - 0.025, n_models_seasonal) * std_prior_SSS_seasonal / np.sqrt(n_models_seasonal),
    mu_prior_SSS_seasonal + stats.t.ppf(1 - 0.025, n_models_seasonal) * std_prior_SSS_seasonal / np.sqrt(n_models_seasonal),
    alpha=0.2, color='g', label='SSS 95% CI'
)

# Add 95% confidence intervals for precipitation
ax2.fill_between(
    seasons_scale,
    mu_prior_precip_seasonal - stats.t.ppf(1 - 0.025, n_models_seasonal) * std_prior_precip_seasonal / np.sqrt(n_models_seasonal),
    mu_prior_precip_seasonal + stats.t.ppf(1 - 0.025, n_models_seasonal) * std_prior_precip_seasonal / np.sqrt(n_models_seasonal),
    alpha=0.2, color='purple', label='Precipitation 95% CI'
)

axes[1].set_title('Prior Mean and 95% Confidence Interval for Seasonal SSS & Precipitation')
axes[1].set_xlabel('Season')
axes[1].set_ylabel('SSS (psu)', color='g')
ax2.set_ylabel('Precipitation (mm/day)', color='purple')
axes[1].set_xticks(seasons_scale)
axes[1].set_xticklabels(seasons)
axes[1].legend(loc='upper left')
ax2.legend(loc='upper right')
axes[1].grid(True)

# Adjust layout and show the plot
plt.tight_layout()
plt.show()

# Set the weights of the data based on the standard errors
weights_seasonal_SST_D47 = 1 / Lutetian_models_seasonal[[f"{season}_SST_D47_SE" for season in seasons]] ** 2
weights_seasonal_SAT_D47 = 1 / Lutetian_models_seasonal[[f"{season}_SAT_D47_SE" for season in seasons]] ** 2

# Change the column suffixes from "_D47_SE" to "_D47" in weights_seasonal to match the headers of the D47 matrix later for multiplication
weights_seasonal_SST_D47.columns = [col.replace('_SST_D47_SE', '_SST_D47') for col in weights_seasonal_SST_D47.columns]
weights_seasonal_SAT_D47.columns = [col.replace('_SAT_D47_SE', '_SAT_D47') for col in weights_seasonal_SAT_D47.columns]

# Prior D47 estimates from climate models (weighted mean)
mu_prior_SST_D47_seasonal = np.array((Lutetian_models_seasonal[[f"{season}_SST_D47" for season in seasons]] * weights_seasonal_SST_D47).sum(axis = 0, skipna = True) / weights_seasonal_SST_D47.sum(axis = 0, skipna = True)) # Calculate weighted seasonal mean D47 values and convert to numpy array
mu_prior_SAT_D47_seasonal = np.array((Lutetian_models_seasonal[[f"{season}_SAT_D47" for season in seasons]] * weights_seasonal_SAT_D47).sum(axis = 0, skipna = True) / weights_seasonal_SAT_D47.sum(axis = 0, skipna = True)) # Calculate weighted seasonal mean D47 values and convert to numpy array

# Calculate simple (unweighted) mean for monthly d18Oc values
mu_prior_d18Oc_seasonal = np.array(Lutetian_models_seasonal[[f"{season}_d18Oc" for season in seasons]].mean(axis=0, skipna=True))
mu_prior_precip_seasonal = np.array(Lutetian_models_seasonal[[f"{season}_precip" for season in seasons]].mean(axis=0, skipna=True))

# Decompose variance within and between model outcomes
model_variances_SST_D47 = Lutetian_models_seasonal[[f"{season}_SST_D47" for season in seasons]].var(axis = 0, ddof = 1)  # Compute variance across models
model_variances_SAT_D47 = Lutetian_models_seasonal[[f"{season}_SAT_D47" for season in seasons]].var(axis = 0, ddof = 1)  # Compute variance across models
measurement_variances_SST_D47 = (Lutetian_models_seasonal[[f"{season}_SST_D47_SE" for season in seasons]] ** 2).mean(axis = 0, skipna = True)  # Compute variance on measurements
measurement_variances_SAT_D47 = (Lutetian_models_seasonal[[f"{season}_SAT_D47_SE" for season in seasons]] ** 2).mean(axis = 0, skipna = True)  # Compute variance on measurements

# Covariance between seasons in prior D47 estimates from climate models (weighted covariance matrix)
cov_raw_seasonal_SST_D47 = np.cov(Lutetian_models_seasonal[[f"{season}_SST_D47" for season in seasons]].dropna(), rowvar = False)  # Compute the covariance matrix for the raw data (without measurement uncertainty)
cov_raw_seasonal_SAT_D47 = np.cov(Lutetian_models_seasonal[[f"{season}_SAT_D47" for season in seasons]].dropna(), rowvar = False)  # Compute the covariance matrix for the raw data (without measurement uncertainty)
cov_raw_seasonal_d18Oc = np.cov(Lutetian_models_seasonal[[f"{season}_d18Oc" for season in seasons]].dropna(), rowvar = False)  # Compute the covariance matrix for the raw data (without measurement uncertainty)
cov_raw_seasonal_precip = np.cov(Lutetian_models_seasonal[[f"{season}_precip" for season in seasons]].dropna(), rowvar = False)  # Compute the covariance matrix for the raw data (without measurement uncertainty)
cov_prior_SST_D47_seasonal = cov_raw_seasonal_SST_D47.copy()  # Copy covariance matrix to add uncertainty coming from the measurements
cov_prior_SAT_D47_seasonal = cov_raw_seasonal_SAT_D47.copy()  # Copy covariance matrix to add uncertainty coming from the measurements
cov_prior_d18Oc_seasonal = cov_raw_seasonal_d18Oc.copy()  # Copy covariance matrix to add uncertainty coming from the measurements
cov_prior_precip_seasonal = cov_raw_seasonal_precip.copy()  # Copy covariance matrix to add uncertainty coming from the measurements
np.fill_diagonal(cov_prior_SST_D47_seasonal, np.diagonal(cov_raw_seasonal_SST_D47) + measurement_variances_SST_D47)  # Add diagonal terms for measurement uncertainties (which have no covariance between models)
np.fill_diagonal(cov_prior_SAT_D47_seasonal, np.diagonal(cov_raw_seasonal_SAT_D47) + measurement_variances_SAT_D47)  # Add diagonal terms for measurement uncertainties (which have no covariance between models)

# Store copy of original prior means to keep when later updating the prior
mu_prior_SST_D47_seasonal_original, cov_prior_SST_D47_seasonal_original = mu_prior_SST_D47_seasonal.copy(), cov_prior_SST_D47_seasonal.copy()
mu_prior_SAT_D47_seasonal_original, cov_prior_SAT_D47_seasonal_original = mu_prior_SAT_D47_seasonal.copy(), cov_prior_SAT_D47_seasonal.copy()
mu_prior_d18Oc_seasonal_original, cov_prior_d18Oc_seasonal_original = mu_prior_d18Oc_seasonal.copy(), cov_raw_seasonal_d18Oc.copy()
mu_prior_precip_seasonal_original, cov_prior_precip_seasonal_original = mu_prior_precip_seasonal.copy(), cov_raw_seasonal_precip.copy()

# Extract the standard deviations (uncertainty) from the covariance matrix
std_prior_SST_D47_seasonal = np.sqrt(np.diag(cov_prior_SST_D47_seasonal))
std_prior_SAT_D47_seasonal = np.sqrt(np.diag(cov_prior_SAT_D47_seasonal))
std_prior_d18Oc_seasonal = np.sqrt(np.diag(cov_prior_d18Oc_seasonal))
std_prior_precip_seasonal = np.sqrt(np.diag(cov_prior_precip_seasonal))

print("mu_prior_SST_D47_seasonal:", mu_prior_SST_D47_seasonal)
print("std_prior_SST_D47_seasonal:", std_prior_SST_D47_seasonal)
print("mu_prior_SAT_D47_seasonal:", mu_prior_SAT_D47_seasonal)
print("std_prior_SAT_D47_seasonal:", std_prior_SAT_D47_seasonal)
print("mu_prior_d18Oc_seasonal:", mu_prior_d18Oc_seasonal)
print("std_prior_d18Oc_seasonal:", std_prior_d18Oc_seasonal)
print("mu_prior_precip_seasonal:", mu_prior_precip_seasonal)
print("std_prior_precip_seasonal:", std_prior_precip_seasonal)

mu_prior_SST_D47_seasonal: [0.59558454 0.5944782  0.5693298  0.57532305]
std_prior_SST_D47_seasonal: [0.00921641 0.00955641 0.00823186 0.00872053]
mu_prior_SAT_D47_seasonal: [0.61833332 0.60060654 0.56627243 0.58983039]
std_prior_SAT_D47_seasonal: [0.01118452 0.01354114 0.00895396 0.01698911]
mu_prior_d18Oc_seasonal: [-0.94610631 -1.04113704 -3.08652898 -2.58441643]
std_prior_d18Oc_seasonal: [0.48936498 0.526327   0.65431916 0.79321889]
mu_prior_precip_seasonal: [0.20279093 0.18348966 0.19218344 0.18032   ]
std_prior_precip_seasonal: [0.06518114 0.07164279 0.10863477 0.06570094]

# Define the seasons, number of models, and scale for the x-axis
seasons = ["winter", "spring", "summer", "autumn"]
n_models_seasonal = len(Lutetian_models["Cell"])  # Find the total number of models (use monthly data because seasonal data has this column duplicated 3 times)
seasons_scale = np.arange(len(seasons)) + 1  # Create seasonal scale

# Create a 2x2 plotting grid
fig, axes = plt.subplots(2, 2, figsize=(15, 12))

# Panel 1: Plot the prior distribution for SST D47 values
axes[0, 0].plot(seasons_scale, mu_prior_SST_D47_seasonal[:len(seasons)], label='Prior SST D47 Mean', marker='o', color='b')
axes[0, 0].fill_between(
    seasons_scale,
    mu_prior_SST_D47_seasonal[:len(seasons)] - stats.t.ppf(1 - 0.025, n_models_seasonal) * std_prior_SST_D47_seasonal / np.sqrt(n_models_seasonal),
    mu_prior_SST_D47_seasonal[:len(seasons)] + stats.t.ppf(1 - 0.025, n_models_seasonal) * std_prior_SST_D47_seasonal / np.sqrt(n_models_seasonal),
    alpha=0.2, color='b', label='95% Confidence Interval'
)
axes[0, 0].set_title('Prior Mean and 95% Confidence Interval for Seasonal SST D47 Values')
axes[0, 0].set_xlabel('Season')
axes[0, 0].set_ylabel('D47 Value')
axes[0, 0].set_xticks(seasons_scale)
axes[0, 0].set_xticklabels(seasons)
axes[0, 0].legend()
axes[0, 0].grid(True)

# Panel 2: Plot the prior distribution for SAT D47 values
axes[0, 1].plot(seasons_scale, mu_prior_SAT_D47_seasonal[:len(seasons)], label='Prior SAT D47 Mean', marker='o', color='r')
axes[0, 1].fill_between(
    seasons_scale,
    mu_prior_SAT_D47_seasonal[:len(seasons)] - stats.t.ppf(1 - 0.025, n_models_seasonal) * std_prior_SAT_D47_seasonal / np.sqrt(n_models_seasonal),
    mu_prior_SAT_D47_seasonal[:len(seasons)] + stats.t.ppf(1 - 0.025, n_models_seasonal) * std_prior_SAT_D47_seasonal / np.sqrt(n_models_seasonal),
    alpha=0.2, color='r', label='95% Confidence Interval'
)
axes[0, 1].set_title('Prior Mean and 95% Confidence Interval for Seasonal SAT D47 Values')
axes[0, 1].set_xlabel('Season')
axes[0, 1].set_ylabel('D47 Value')
axes[0, 1].set_xticks(seasons_scale)
axes[0, 1].set_xticklabels(seasons)
axes[0, 1].legend()
axes[0, 1].grid(True)

# Panel 3: Plot the prior distribution for d18Oc
axes[1, 0].plot(seasons_scale, mu_prior_d18Oc_seasonal, label='Prior d18Oc Mean', marker='o', color='g')
axes[1, 0].fill_between(
    seasons_scale,
    mu_prior_d18Oc_seasonal - stats.t.ppf(1 - 0.025, n_models_seasonal) * std_prior_d18Oc_seasonal / np.sqrt(n_models_seasonal),
    mu_prior_d18Oc_seasonal + stats.t.ppf(1 - 0.025, n_models_seasonal) * std_prior_d18Oc_seasonal / np.sqrt(n_models_seasonal),
    alpha=0.2, color='g', label='95% Confidence Interval'
)
axes[1, 0].set_title('Prior Mean and 95% Confidence Interval for Seasonal d18Oc Values')
axes[1, 0].set_xlabel('Season')
axes[1, 0].set_ylabel('d18Oc Value')
axes[1, 0].set_xticks(seasons_scale)
axes[1, 0].set_xticklabels(seasons)
axes[1, 0].legend()
axes[1, 0].grid(True)

# Panel 4: Plot the prior distribution for precipitation
axes[1, 1].plot(seasons_scale, mu_prior_precip_seasonal, label='Prior Precipitation Mean', marker='o', color='purple')
axes[1, 1].fill_between(
    seasons_scale,
    mu_prior_precip_seasonal - stats.t.ppf(1 - 0.025, n_models_seasonal) * std_prior_precip_seasonal / np.sqrt(n_models_seasonal),
    mu_prior_precip_seasonal + stats.t.ppf(1 - 0.025, n_models_seasonal) * std_prior_precip_seasonal / np.sqrt(n_models_seasonal),
    alpha=0.2, color='purple', label='95% Confidence Interval'
)
axes[1, 1].set_title('Prior Mean and 95% Confidence Interval for Seasonal Precipitation Values')
axes[1, 1].set_xlabel('Season')
axes[1, 1].set_ylabel('Precipitation (mm/day)')
axes[1, 1].set_xticks(seasons_scale)
axes[1, 1].set_xticklabels(seasons)
axes[1, 1].legend()
axes[1, 1].grid(True)

# Adjust layout and show the plot
plt.tight_layout()
plt.show()

# Extract the relevant columns for SST, SAT D47, d18Oc, and precipitation
SST_D47_columns_seasonal = [f"{season}_SST_D47" for season in seasons]
SAT_D47_columns_seasonal = [f"{season}_SAT_D47" for season in seasons]
d18Oc_columns_seasonal = [f"{season}_d18Oc" for season in seasons]
precip_columns_seasonal = [f"{season}_precip" for season in seasons]

# Combine the relevant columns into a single dataframe
combined_data_seasonal = Lutetian_models_seasonal[
    SST_D47_columns_seasonal + SAT_D47_columns_seasonal + d18Oc_columns_seasonal + precip_columns_seasonal
]

# Calculate the covariance matrix for the combined data
cov_combined_seasonal = np.cov(combined_data_seasonal.dropna(), rowvar=False)

# Extract the covariance matrices for each variable
cov_SST_D47_seasonal = cov_combined_seasonal[:len(seasons), :len(seasons)]
cov_SAT_D47_seasonal = cov_combined_seasonal[len(seasons):2*len(seasons), len(seasons):2*len(seasons)]
cov_d18Oc_seasonal = cov_combined_seasonal[2*len(seasons):3*len(seasons), 2*len(seasons):3*len(seasons)]
cov_precip_seasonal = cov_combined_seasonal[3*len(seasons):, 3*len(seasons):]

# Extract the cross-covariance matrices
cross_cov_SST_SAT_D47_seasonal = cov_combined_seasonal[:len(seasons), len(seasons):2*len(seasons)]
cross_cov_SST_d18Oc_seasonal = cov_combined_seasonal[:len(seasons), 2*len(seasons):3*len(seasons)]
cross_cov_SST_precip_seasonal = cov_combined_seasonal[:len(seasons), 3*len(seasons):]
cross_cov_SAT_d18Oc_seasonal = cov_combined_seasonal[len(seasons):2*len(seasons), 2*len(seasons):3*len(seasons)]
cross_cov_SAT_precip_seasonal = cov_combined_seasonal[len(seasons):2*len(seasons), 3*len(seasons):]
cross_cov_d18Oc_precip_seasonal = cov_combined_seasonal[2*len(seasons):3*len(seasons), 3*len(seasons):]

# Plot a heatmap of the combined covariance matrix
plt.figure(figsize=(12, 10))
sns.heatmap(
    np.round(cov_combined_seasonal * 10**4, 1),  # Scale by 10^4 for better visualization and round values
    annot=False,
    fmt=".2f",
    cmap="coolwarm",
    center=0,
    xticklabels=SST_D47_columns_seasonal + SAT_D47_columns_seasonal + d18Oc_columns_seasonal + precip_columns_seasonal,
    yticklabels=SST_D47_columns_seasonal + SAT_D47_columns_seasonal + d18Oc_columns_seasonal + precip_columns_seasonal
)

# Add titles to the axes per parameter
plt.axvline(x=len(SST_D47_columns_seasonal), color='black', linestyle='--', linewidth=1)
plt.axvline(x=len(SST_D47_columns_seasonal) + len(SAT_D47_columns_seasonal), color='black', linestyle='--', linewidth=1)
plt.axvline(x=len(SST_D47_columns_seasonal) + len(SAT_D47_columns_seasonal) + len(d18Oc_columns_seasonal), color='black', linestyle='--', linewidth=1)

plt.axhline(y=len(SST_D47_columns_seasonal), color='black', linestyle='--', linewidth=1)
plt.axhline(y=len(SST_D47_columns_seasonal) + len(SAT_D47_columns_seasonal), color='black', linestyle='--', linewidth=1)
plt.axhline(y=len(SST_D47_columns_seasonal) + len(SAT_D47_columns_seasonal) + len(d18Oc_columns_seasonal), color='black', linestyle='--', linewidth=1)

# Add parameter labels
plt.text(len(SST_D47_columns_seasonal) / 2, -2, 'SST D47', ha='center', va='center', fontsize=10)
plt.text(len(SST_D47_columns_seasonal) + len(SAT_D47_columns_seasonal) / 2, -2, 'SAT D47', ha='center', va='center', fontsize=10)
plt.text(len(SST_D47_columns_seasonal) + len(SAT_D47_columns_seasonal) + len(d18Oc_columns_seasonal) / 2, -2, 'd18Oc', ha='center', va='center', fontsize=10)
plt.text(len(SST_D47_columns_seasonal) + len(SAT_D47_columns_seasonal) + len(d18Oc_columns_seasonal) + len(precip_columns_seasonal) / 2, -2, 'Precipitation', ha='center', va='center', fontsize=10)

plt.title("Combined Covariance Matrix for SST D47, SAT D47, d18Oc, and Precipitation")
plt.show()

# Normalize each submatrix independently for better visualization
def normalize_matrix(matrix):
    min_val = np.min(matrix)
    max_val = np.max(matrix)
    return (matrix - min_val) / (max_val - min_val)

# Extract the covariance matrices for SAT D47, SST D47, d18Oc, and precipitation
cov_SAT_D47_seasonal = cov_combined_seasonal[:len(seasons), :len(seasons)]
cov_SST_D47_seasonal = cov_combined_seasonal[len(seasons):2*len(seasons), len(seasons):2*len(seasons)]
cov_d18Oc_seasonal = cov_combined_seasonal[2*len(seasons):3*len(seasons), 2*len(seasons):3*len(seasons)]
cov_precip_seasonal = cov_combined_seasonal[3*len(seasons):, 3*len(seasons):]

# Extract the cross-covariance matrices
cross_cov_SAT_SST_D47_seasonal = cov_combined_seasonal[:len(seasons), len(seasons):2*len(seasons)]
cross_cov_SAT_d18Oc_seasonal = cov_combined_seasonal[:len(seasons), 2*len(seasons):3*len(seasons)]
cross_cov_SAT_precip_seasonal = cov_combined_seasonal[:len(seasons), 3*len(seasons):]
cross_cov_SST_d18Oc_seasonal = cov_combined_seasonal[len(seasons):2*len(seasons), 2*len(seasons):3*len(seasons)]
cross_cov_SST_precip_seasonal = cov_combined_seasonal[len(seasons):2*len(seasons), 3*len(seasons):]
cross_cov_d18Oc_precip_seasonal = cov_combined_seasonal[2*len(seasons):3*len(seasons), 3*len(seasons):]

# Normalize each submatrix
normalized_cov_SAT_D47_seasonal = normalize_matrix(cov_SAT_D47_seasonal)
normalized_cov_SST_D47_seasonal = normalize_matrix(cov_SST_D47_seasonal)
normalized_cov_d18Oc_seasonal = normalize_matrix(cov_d18Oc_seasonal)
normalized_cov_precip_seasonal = normalize_matrix(cov_precip_seasonal)

# Normalize each cross-covariance matrix
normalized_cross_cov_SAT_SST_D47_seasonal = normalize_matrix(cross_cov_SAT_SST_D47_seasonal)
normalized_cross_cov_SAT_d18Oc_seasonal = normalize_matrix(cross_cov_SAT_d18Oc_seasonal)
normalized_cross_cov_SAT_precip_seasonal = normalize_matrix(cross_cov_SAT_precip_seasonal)
normalized_cross_cov_SST_d18Oc_seasonal = normalize_matrix(cross_cov_SST_d18Oc_seasonal)
normalized_cross_cov_SST_precip_seasonal = normalize_matrix(cross_cov_SST_precip_seasonal)
normalized_cross_cov_d18Oc_precip_seasonal = normalize_matrix(cross_cov_d18Oc_precip_seasonal)

# Combine the normalized submatrices into a single normalized covariance matrix
normalized_cov_combined_seasonal = np.block([
    [normalized_cov_SAT_D47_seasonal, normalized_cross_cov_SAT_SST_D47_seasonal, normalized_cross_cov_SAT_d18Oc_seasonal, normalized_cross_cov_SAT_precip_seasonal],
    [normalized_cross_cov_SAT_SST_D47_seasonal.T, normalized_cov_SST_D47_seasonal, normalized_cross_cov_SST_d18Oc_seasonal, normalized_cross_cov_SST_precip_seasonal],
    [normalized_cross_cov_SAT_d18Oc_seasonal.T, normalized_cross_cov_SST_d18Oc_seasonal.T, normalized_cov_d18Oc_seasonal, normalized_cross_cov_d18Oc_precip_seasonal],
    [normalized_cross_cov_SAT_precip_seasonal.T, normalized_cross_cov_SST_precip_seasonal.T, normalized_cross_cov_d18Oc_precip_seasonal.T, normalized_cov_precip_seasonal]
])

# Plot the heatmap of the normalized combined covariance matrix
plt.figure(figsize=(12, 10))
sns.heatmap(
    normalized_cov_combined_seasonal,
    annot=False,
    fmt=".2f",
    cmap="coolwarm",
    center=0,
    xticklabels=SAT_D47_columns_seasonal + SST_D47_columns_seasonal + d18Oc_columns_seasonal + precip_columns_seasonal,
    yticklabels=SAT_D47_columns_seasonal + SST_D47_columns_seasonal + d18Oc_columns_seasonal + precip_columns_seasonal
)

# Add titles to the axes per parameter
plt.axvline(x=len(SAT_D47_columns_seasonal), color='black', linestyle='--', linewidth=1)
plt.axvline(x=len(SAT_D47_columns_seasonal) + len(SST_D47_columns_seasonal), color='black', linestyle='--', linewidth=1)
plt.axvline(x=len(SAT_D47_columns_seasonal) + len(SST_D47_columns_seasonal) + len(d18Oc_columns_seasonal), color='black', linestyle='--', linewidth=1)

plt.axhline(y=len(SAT_D47_columns_seasonal), color='black', linestyle='--', linewidth=1)
plt.axhline(y=len(SAT_D47_columns_seasonal) + len(SST_D47_columns_seasonal), color='black', linestyle='--', linewidth=1)
plt.axhline(y=len(SAT_D47_columns_seasonal) + len(SST_D47_columns_seasonal) + len(d18Oc_columns_seasonal), color='black', linestyle='--', linewidth=1)

# Add parameter labels
plt.text(len(SAT_D47_columns_seasonal) / 2, -1, 'D47 value from SAT', ha='center', va='center', fontsize=10)
plt.text(len(SAT_D47_columns_seasonal) + len(SST_D47_columns_seasonal) / 2, -1, 'D47 value from SST', ha='center', va='center', fontsize=10)
plt.text(len(SAT_D47_columns_seasonal) + len(SST_D47_columns_seasonal) + len(d18Oc_columns_seasonal) / 2, -1, 'd18Oc', ha='center', va='center', fontsize=10)
plt.text(len(SAT_D47_columns_seasonal) + len(SST_D47_columns_seasonal) + len(d18Oc_columns_seasonal) + len(precip_columns_seasonal) / 2, -1, 'Precipitation', ha='center', va='center', fontsize=10)

plt.text(-3, len(SAT_D47_columns_seasonal) / 2, 'D47 value from SAT', ha='center', va='center', rotation=90, fontsize=10)
plt.text(-3, len(SAT_D47_columns_seasonal) + len(SST_D47_columns_seasonal) / 2, 'D47 value from SST', ha='center', va='center', rotation=90, fontsize=10)
plt.text(-3, len(SAT_D47_columns_seasonal) + len(SST_D47_columns_seasonal) + len(d18Oc_columns_seasonal) / 2, 'd18Oc', ha='center', va='center', rotation=90, fontsize=10)
plt.text(-3, len(SAT_D47_columns_seasonal) + len(SST_D47_columns_seasonal) + len(d18Oc_columns_seasonal) + len(precip_columns_seasonal) / 2, 'Precipitation', ha='center', va='center', rotation=90, fontsize=10)

plt.title("Normalized Combined Covariance Matrix")
plt.show()

# Combine the prior means of D47 and SAT into a single state vector
mu_prior_seasonal_combined = np.concatenate((mu_prior_SST_D47_seasonal, mu_prior_SAT_D47_seasonal, mu_prior_d18Oc_seasonal, mu_prior_precip_seasonal))

# Combine the covariance matrices of D47 values of SST and SAT, d18Oc and precipitation including the cross-covariance
cov_prior_seasonal_combined = cov_combined_seasonal.copy()

# Load seasonal measurements and format them into a dictionary
# This is precompiled seasonal data per specimen and therefore does not come with a time uncertainty
Lutetian_seasonally_aggregated_data = pd.read_csv('Lutetian case/D47_season_data_calc.csv') # Load the data for seasonal averages
Lutetian_seasonally_aggregated_data_dict = Lutetian_seasonally_aggregated_data.to_dict('records') # Convert to dictionary with column headers as keys

# Add an entry for the time uncertainty (which is always zero in this case, because data is already aggregated monthly)
for record in Lutetian_seasonally_aggregated_data_dict:
    record["Season_err"] = 0 # Set the time uncertainty to zero
    record["D47_se"] = record["D47_SD"] / np.sqrt(record["count"]) # Calculate the standard error of the D47 value

print(Lutetian_seasonally_aggregated_data_dict[0]) # Print to check the structure of the data

{'Season': 'summer', 'Whorl...P.or.T': 'AW', 'X': 'AX', 'D47_mean': 0.589860492, 'D47_SD': 0.042653301, 'count': 62, 'd18O': -2.150527044, 'd18O_SD': 0.312787802, 'T': 27.892776, 'CL95': 0.010920335, 'CL95_T': 3.824099323, 'd18Osw': -0.27016391, 'DOY': 203.6572, 'Tmin': 24.20918351, 'Tmax': 31.71674359, 'dwmin': -1.11891794, 'dwmax': 0.610934613, 'Season_err': 0, 'D47_se': 0.005416974643977352}

# Load measurements and format them into a dictionary
# These are the actual individual D47 and d18Oc measurements and ShellChron outcomes and thus come with a time uncertainty which can be propagated.
Lutetian_D47_data = pd.read_csv('Lutetian case/Campanile_sample_data_calc2.csv') # Load data for individual D47 and d18Oc measurements and ShellChron outcomes
Lutetian_data_dict = Lutetian_D47_data.to_dict('records') # Convert to dictionary with column headers as keys

# Add an entry for the time uncertainty (which is always zero in this case, because there is no time uncertainty in the raw data)
for record in Lutetian_data_dict:
    # Handle missing values and convert from days to months and seasons
    shell_chron_doy_err = record.get("DOY_SD", np.nan)  # Get value, default to NaN if missing
    if pd.isna(shell_chron_doy_err):  # Check if the value is NaN
        record["Month_err"] = 0 # Set the time uncertainty to zero
        record["Season_err"] = 0 # Set the time uncertainty to zero
        record["no_err"] = 0 # Set the time uncertainty to zero
    else:
        record["Month_err"] = shell_chron_doy_err / 365 * 12  # Convert days to months
        record["Season_err"] = shell_chron_doy_err / 365 * 4 # Convert days to seasons
        record["no_err"] = 0 # Set the time uncertainty to zero for no error
    # Assign the D47 value and its standard deviation
    record["D47_SD"] = 0.029 # Assign external standard deviation to the D47 value (based on reproducibility of IAEA-C2 measurements)
    record["season_score"] = np.floor(((record["DOY"] + 45) % 365) / 365 * 4) # Calculate the season score from DOY value (winter = 0, spring = 1, summer = 2, autumn = 3). Shift two months to align with the seasons defined in paper.
    record["month_score"] = np.floor(((record["DOY"]) % 365) / 365 * 12) # Calculate the month score from DOY value (January = 0, February = 1, ..., December = 11).

print(Lutetian_data_dict[0]) # Print to check the structure of the data

{'ID': 'AB002', 'D': 5.0, 'Run': nan, 'Row': nan, 'Sample.intensity': nan, 'X49.parameter': nan, 'D47_raw': nan, 'D47_SD': 0.029, 'D47_final': nan, 'Temperature': nan, 'd18O': -1.5, 'd13C': 2.15, 'Whorl': nan, 'sample_nr': nan, 'Year': nan, 'season_manual': nan, 'd18O_SD': 0.1, 'day': 145.316271, 'DOY_SD': nan, 'season_label': nan, 'T_d18O': nan, 'DOY': 145.316271, 'Season': nan, 'Month_err': 0, 'Season_err': 0, 'no_err': 0, 'season_score': 2.0, 'month_score': 4.0}

# Define seasonal IDs based on season names
season_names = {
    0: "winter",
    1: "spring",
    2: "summer",
    3: "autumn"
}

# Add seasonal and monthly ID to the proxy data
for record in Lutetian_data_dict:
    # Attach the correct season name based on season_score
    season = int(record["season_score"]) if not pd.isna(record["season_score"]) else None
    if season is not None and season in season_names:
        record["season_name"] = season_names[season]
    else:
        record["season_name"] = None

for record in Lutetian_seasonally_aggregated_data_dict:
    record["Season_err"] = 0 # Set the season error to zero
    for season_score, season_name in season_names.items():
        if record["Season"] == season_name:
            record["season_score"] = season_score
            break

# Aggregate and summarize proxy data per month, tracking mean and propagated SE for D47 and d18O separately
monthly_agg = []
for month in range(12):
    month_records = [record for record in Lutetian_data_dict if int(record["month_score"]) == month]
    entry = {
        "month_score": month,
        "D47_mean": np.nanmean([rec.get("D47_final") for rec in month_records if not pd.isna(rec.get("D47_final"))]),
        "D47_SD": np.sqrt(np.nansum([rec.get("D47_SD", np.nan) ** 2 for rec in month_records if not pd.isna(rec.get("D47_SD")) and not pd.isna(rec.get("D47_final"))])) / max(1, len([rec for rec in month_records if not pd.isna(rec.get("D47_SD")) and not pd.isna(rec.get("D47_final"))])),
        "d18O": np.nanmean([rec.get("d18O") for rec in month_records if not pd.isna(rec.get("d18O"))]),
        "d18O_SD": np.sqrt(np.nansum([rec.get("d18O_SD", np.nan) ** 2 for rec in month_records if not pd.isna(rec.get("d18O_SD")) and not pd.isna(rec.get("d18O"))])) / max(1, len([rec for rec in month_records if not pd.isna(rec.get("d18O_SD")) and not pd.isna(rec.get("d18O"))])),
        "Month_err": 0
    }
    monthly_agg.append(entry)
Lutetian_monthly_aggregated_data_df = pd.DataFrame(monthly_agg)
Lutetian_monthly_aggregated_data_dict = Lutetian_monthly_aggregated_data_df.to_dict("records")

print(Lutetian_data_dict[0]) # Print to check the structure of the data
print(Lutetian_seasonally_aggregated_data_dict[0]) # Print to check the structure of the data
print(Lutetian_monthly_aggregated_data_dict[0]) # Print to check the structure of the data

{'ID': 'AB002', 'D': 5.0, 'Run': nan, 'Row': nan, 'Sample.intensity': nan, 'X49.parameter': nan, 'D47_raw': nan, 'D47_SD': 0.029, 'D47_final': nan, 'Temperature': nan, 'd18O': -1.5, 'd13C': 2.15, 'Whorl': nan, 'sample_nr': nan, 'Year': nan, 'season_manual': nan, 'd18O_SD': 0.1, 'day': 145.316271, 'DOY_SD': nan, 'season_label': nan, 'T_d18O': nan, 'DOY': 145.316271, 'Season': nan, 'Month_err': 0, 'Season_err': 0, 'no_err': 0, 'season_score': 2.0, 'month_score': 4.0, 'season_name': 'summer'}
{'Season': 'summer', 'Whorl...P.or.T': 'AW', 'X': 'AX', 'D47_mean': 0.589860492, 'D47_SD': 0.042653301, 'count': 62, 'd18O': -2.150527044, 'd18O_SD': 0.312787802, 'T': 27.892776, 'CL95': 0.010920335, 'CL95_T': 3.824099323, 'd18Osw': -0.27016391, 'DOY': 203.6572, 'Tmin': 24.20918351, 'Tmax': 31.71674359, 'dwmin': -1.11891794, 'dwmax': 0.610934613, 'Season_err': 0, 'D47_se': 0.005416974643977352, 'season_score': 2}
{'month_score': 0, 'D47_mean': 0.6315000000000001, 'D47_SD': 0.0145, 'd18O': 0.016000000000000014, 'd18O_SD': 0.0447213595499958, 'Month_err': 0}

C:\Users\nwi213\AppData\Local\Temp\ipykernel_131304\4127643076.py:31: RuntimeWarning: Mean of empty slice
  "D47_mean": np.nanmean([rec.get("D47_final") for rec in month_records if not pd.isna(rec.get("D47_final"))]),

# Function to calculate wrapped normal distribution weights
def wrapped_normal_pdf(x, mean, sd, num_bins):
    # Calculate the normal PDF for each bin
    pdf = stats.norm.pdf(x, loc = mean, scale = sd)
    # Wrap around the bins
    for i in range(1, num_bins):
        pdf += stats.norm.pdf(x + i * num_bins, loc = mean, scale = sd)
        pdf += stats.norm.pdf(x - i * num_bins, loc = mean, scale = sd)
    # Normalize the weights to ensure the sum equals 1
    pdf /= pdf.sum()
    return pdf

# Extract measurements and uncertainties from the dictionary
D47_measurements_seasonal_aggregated = [measurement["D47_mean"] for measurement in Lutetian_seasonally_aggregated_data_dict] # Extract the D47 values
D47_measurements_monthly_aggregated = [measurement["D47_mean"] for measurement in Lutetian_monthly_aggregated_data_dict] # Extract the D47 values from monthly aggregated data
d18Oc_measurements_seasonal_aggregated = [measurement["d18O"] for measurement in Lutetian_seasonally_aggregated_data_dict] # Extract the d18Oc values
d18Oc_measurements_monthly_aggregated = [measurement["d18O"] for measurement in Lutetian_monthly_aggregated_data_dict] # Extract the d18Oc values from monthly aggregated data
D47_measurements_seasonal_aggregated_se = [measurement["D47_se"] ** 2 for measurement in Lutetian_seasonally_aggregated_data_dict] # Square the standard deviation to get the variance
D47_measurements_monthly_aggregated_se = [measurement["D47_SD"] ** 2 for measurement in Lutetian_monthly_aggregated_data_dict] # Square the standard deviation to get the variance
d18Oc_measurements_seasonal_aggregated_se = [measurement["d18O_SD"] ** 2 for measurement in Lutetian_seasonally_aggregated_data_dict] # Square the standard deviation to get the variance
d18Oc_measurements_monthly_aggregated_se = [measurement["d18O_SD"] ** 2 for measurement in Lutetian_monthly_aggregated_data_dict] # Square the standard deviation to get the variance

# OPTIONAL: Lower boundary d18Oc variance at 0.01 (equivalent to 0.1 per mil measurement uncertainty)
d18Oc_measurements_seasonal_aggregated_se = [max(uncertainty, 0.01) for uncertainty in d18Oc_measurements_seasonal_aggregated_se]  # Ensure the uncertainty variance is at least 0.01
d18Oc_measurements_monthly_aggregated_se = [max(uncertainty, 0.01) for uncertainty in d18Oc_measurements_monthly_aggregated_se]  # Ensure the uncertainty variance is at least 0.01

# Create the measurement matrix Z
Z_seasonal_aggregated = np.array(D47_measurements_seasonal_aggregated + d18Oc_measurements_seasonal_aggregated).reshape(-1, 1) # Combine D47 and d18Oc measurements into a single matrix
Z_monthly_aggregated = np.array(D47_measurements_monthly_aggregated + d18Oc_measurements_monthly_aggregated).reshape(-1, 1) # Combine D47 and d18Oc measurements into a single matrix

# Create the measurement uncertainty matrix R (diagonal matrix)
R_seasonal_aggregated = np.diag(D47_measurements_seasonal_aggregated_se + d18Oc_measurements_seasonal_aggregated_se) # Combine the variances of D47 and d18Oc measurements into a single diagonal matrix
R_monthly_aggregated = np.diag(D47_measurements_monthly_aggregated_se + d18Oc_measurements_monthly_aggregated_se) # Combine the variances of D47 and d18Oc measurements into a single diagonal matrix

# # If NaN values are present in the measurements, remove them as well as the corresponding uncertainties
# R_seasonal_aggregated = R_seasonal_aggregated[~np.isnan(Z_seasonal_aggregated).any(axis=1), :][:, ~np.isnan(Z_seasonal_aggregated).any(axis=1)]  # Remove rows and columns with NaN values
# R_monthly_aggregated = R_monthly_aggregated[~np.isnan(Z_monthly_aggregated).any(axis=1), :][:, ~np.isnan(Z_monthly_aggregated).any(axis=1)]  # Remove rows and columns with NaN values
# Z_seasonal_aggregated = Z_seasonal_aggregated[~np.isnan(Z_seasonal_aggregated).any(axis=1)]  # Remove rows with NaN values
# Z_monthly_aggregated = Z_monthly_aggregated[~np.isnan(Z_monthly_aggregated).any(axis=1)]  # Remove rows with NaN values

# Number of seasonally averaged measurements
N_measurements_seasonal_aggregated = len(Z_seasonal_aggregated) # Get the number of aggregated measurements
N_measurements_monthly_aggregated = len(Z_monthly_aggregated) # Get the number of aggregated measurements

# Create the observation matrix H for seasonal data based on seasonally aggregated data
H_seasonal_aggregated = np.zeros((N_measurements_seasonal_aggregated, len(mu_prior_seasonal_combined)))
H_monthly_aggregated = np.zeros((N_measurements_monthly_aggregated, len(mu_prior_monthly_combined)))

# Fill the seasonal observation matrix H with ones at the positions corresponding to the measurements
half_seasonal = int(N_measurements_seasonal_aggregated / 2)
for i, measurement in enumerate(Lutetian_seasonally_aggregated_data_dict):
    season_index = int(measurement["season_score"])
    # First half: D47 (SST)
    H_seasonal_aggregated[i, season_index] = 1
    # Second half: d18Oc (SSS)
    H_seasonal_aggregated[i + half_seasonal, season_index + 8] = 1

# Fill the monthly observation matrix H with ones at the positions corresponding to the measurements
half_monthly = int(N_measurements_monthly_aggregated / 2)
for i, measurement in enumerate(Lutetian_monthly_aggregated_data_dict):
    month_index = int(measurement["month_score"])
    # First half: D47 (SST)
    H_monthly_aggregated[i, month_index] = 1
    # Second half: d18Oc (SSS)
    H_monthly_aggregated[i + half_monthly, month_index + 24] = 1

print(H_seasonal_aggregated)
print(H_monthly_aggregated)

[[0. 0. 1. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
 [0. 0. 0. 1. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
 [1. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
 [0. 1. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
 [0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 1. 0. 0. 0. 0. 0.]
 [0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 1. 0. 0. 0. 0.]
 [0. 0. 0. 0. 0. 0. 0. 0. 1. 0. 0. 0. 0. 0. 0. 0.]
 [0. 0. 0. 0. 0. 0. 0. 0. 0. 1. 0. 0. 0. 0. 0. 0.]]
[[1. 0. 0. ... 0. 0. 0.]
 [0. 1. 0. ... 0. 0. 0.]
 [0. 0. 1. ... 0. 0. 0.]
 ...
 [0. 0. 0. ... 0. 0. 0.]
 [0. 0. 0. ... 0. 0. 0.]
 [0. 0. 0. ... 0. 0. 0.]]

def kalman_update_block(
    mu_prior: np.ndarray,
    cov_prior: np.ndarray,
    Z: np.ndarray,
    R: np.ndarray,
    H: np.ndarray,
    debug_print: bool = False
):
    """
    Perform a Kalman update step for a block of observations.

    Parameters:
    mu_prior (np.ndarray): The prior mean vector.
    cov_prior (np.ndarray): The prior covariance matrix.
    Z (np.ndarray): The measurement matrix.
    R (np.ndarray): The measurement noise covariance matrix.
    H (np.ndarray): The observation matrix.
    debug_print (bool): If True, print debug statements.

    Returns:
    mu_posterior (np.ndarray): The posterior mean vector.
    cov_posterior (np.ndarray): The posterior covariance matrix.
    """
    if debug_print:
        # Print shapes of key variables for debugging
        print("Shape of cov_prior:", cov_prior.shape)
        print("Shape of H:", H.shape)
        print("Shape of R:", R.shape)
        print("Shape of mu_prior:", mu_prior.shape)
        print("Shape of Z:", Z.shape)

    # Compute the Kalman gain
    K = cov_prior @ H.T @ np.linalg.inv(H @ cov_prior @ H.T + R)
    if debug_print:
        print("K matrix:", K)
        print("Shape of K:", K.shape)

    # In-between steps for debugging
    Y_hat = H @ mu_prior  # Compute the predicted observation
    if debug_print:
        print("Y_hat:", Y_hat)
        print("Shape of Y_hat:", Y_hat.shape)
    innovation = Z - Y_hat.reshape(-1, 1)  # Compute the innovation
    if debug_print:
        print("Innovation:", innovation)
        print("Shape of innovation:", innovation.shape)
    # Replace NaN values in innovation with zeros
    innovation = np.nan_to_num(innovation, nan=0.0)
    kalman_gain = (K @ innovation).flatten()
    if debug_print:
        print("Kalman gain:", kalman_gain)
        print("Shape of kalman_gain:", kalman_gain.shape)

    # Update the posterior mean estimate
    mu_posterior = mu_prior + kalman_gain.flatten()

    # Update the posterior covariance estimate
    cov_posterior = cov_prior - K @ H @ cov_prior

    return mu_posterior, cov_posterior

# UPDATED SCRIPT TO ACCOMMODATE MULTIPLE VARIABLES
# Create function to keep track of the likelihood statistics and data

def likelihood_statistics_multi(
    weighted_sum,
    effective_weights_total,
    n_update,
    data_library,
    measurement,
    timestamp,
    timestamp_sd,
    Variable_names = ["Variable_name1", "Variable_name2"],
    Variable_names_SDs = ["Variable_name_SD1", "Variable_name_SD2"]
):
    """
    Incrementally updates the likelihood statistics for seasonal data.

    Parameters:
    - weighted_sum: list
        List tracking the mean times the effective weight for each time bin and variable.
    - effective_weights_total: list
        List tracking the sum of effective weights for each time bin and variable.
    - n_update: list
        List tracking the number of datapoints for each time bin and variable.
    - data_library: dict
        Dictionary tracking individual data points and their uncertainties.
    - measurement: dict
        A single measurement containing data on multiple variables.
    - timestamp: str
        Key in the measurement dictionary for the timestamp (0-based index).
    - timestamp_sd: float
        Standard deviation of uncertainty in the timestamp.
    - Variable_name: list of str
        Key in the measurement dictionary for the variables (e.g. d18Os, D47).
    - Variable_name_SD: list of str
        Key in the measurement dictionary for the standard deviation on the variables (e.g. d18Os, D47).
    """
    # Check if at least one combination of variable name and its SD is present in the measurement
    found = False
    for var_name, var_sd_name in zip(Variable_names, Variable_names_SDs):
        if var_name in measurement and var_sd_name in measurement:
            found = True
            break

    if timestamp in measurement and found:
        # Extract the time and data values from the measurement
        time = measurement[timestamp]
        time_sd = measurement[timestamp_sd]
        # Loop through all variable/SD pairs
        for var_name, var_sd_name in zip(Variable_names, Variable_names_SDs):
            if var_name in measurement and var_sd_name in measurement:
                data_val = measurement[var_name]
                data_sd = measurement[var_sd_name]
                
                # Check if the data is valid
                if not np.isnan(data_val) and not np.isnan(data_sd):
                    # Calculate the weight (inverse of variance)
                    weight = 1 / (data_sd ** 2)

                    # Determine the number of bins
                    num_bins_seasonal = int(len(weighted_sum) / len(Variable_names))
                    # Ensure num_bins_seasonal is an integer
                    bin_indices = np.arange(num_bins_seasonal, dtype=np.float64)

                    # Calculate the probability density for each bin
                    if time_sd == 0:  # Catch cases where the time uncertainty is zero (or unknown)
                        probabilities = np.zeros(num_bins_seasonal, dtype=np.float64)
                        bin_index = int(time) % num_bins_seasonal  # Ensure the bin index is within range
                        probabilities[bin_index] = 1  # Set the probability to 1 for the correct bin
                    else:
                        probabilities = stats.norm.pdf(bin_indices, loc=time, scale=time_sd)  # For non-zero time uncertainty, use a normal distribution
                        probabilities /= probabilities.sum()  # Normalize to ensure the sum of probabilities is 1

                    for i, prob in enumerate(probabilities):  # Loop over all possible bin numbers in the probability vector
                        bin_index = i % num_bins_seasonal  # Wrap around to the first bin if it overflows

                        # Update the weighted sums and sample count
                        effective_weight = weight * prob
                        var_idx = Variable_names.index(var_name)  # Find the index of the variable
                        idx = int(var_idx * num_bins_seasonal + bin_index)  # Unique index for (variable, bin)
                        if weighted_sum[idx] is None:
                            weighted_sum[idx] = 0
                            effective_weights_total[idx] = 0
                        weighted_sum[idx] = weighted_sum[idx] + data_val * effective_weight
                        effective_weights_total[idx] = effective_weights_total[idx] + effective_weight

                    # Update n_update for the correct variable and bin
                    var_idx = Variable_names.index(var_name) # Find the index of the variable
                    n_update[var_idx * num_bins_seasonal + (int(time) % num_bins_seasonal)] += 1  # update sample number per bin and variable

                    # Track individual data points and their uncertainties
                    key = (var_name, int(time)) # Store individual data points in a dictionary with (variable, time) as key
                    if key not in data_library:
                        data_library[key] = []  # Initialize the list for a new (time, var_name) pair
                    data_library[key].append((time_sd, data_val, data_sd))
    return weighted_sum, effective_weights_total, n_update, data_library # Return the updated values

# Apply Kalman function to update the prior with seasonal data including updating the prec estimates
# Update the seasonal D47 and prec prior with all measurements using block updating
seasonal_aggregated_data = {} # Keep track of datapoints per season
n_update_seasonal_aggregated = np.concatenate([mu_prior_SST_D47_seasonal * 0, mu_prior_d18Oc_seasonal * 0]) # Vector to store sample size per season for confidence interval plotting
weighted_sum_seasonal_aggregated = np.concatenate([mu_prior_SST_D47_seasonal * 0, mu_prior_d18Oc_seasonal * 0]) # Vector to store mean temperature per season for confidence interval plotting
effective_weights_total_seasonal_aggregated = np.concatenate([mu_prior_SST_D47_seasonal * 0, mu_prior_d18Oc_seasonal * 0]) # Vector to store temperature uncertainty per season for confidence interval plotting
mu_likelihood_seasonal_aggregated = np.concatenate([mu_prior_SST_D47_seasonal * 0, mu_prior_d18Oc_seasonal * 0]) # Vector to store mean temperature per season for confidence interval plotting
std_likelihood_seasonal_aggregated = np.concatenate([mu_prior_SST_D47_seasonal * 0, mu_prior_d18Oc_seasonal * 0]) # Vector to store temperature uncertainty per season for confidence interval plotting
var_names = ["D47_mean", "d18O"] # List of variable names which are updated
var_SD_names = ["D47_SD", "d18O_SD"] # List of names of variable uncertainties which are updated

# Update the prior with seasonal data using the Kalman filter in block updating form
mu_post_seasonal_aggregated, cov_post_seasonal_aggregated = kalman_update_block(
    mu_prior_seasonal_combined,
    cov_prior_seasonal_combined,
    Z_seasonal_aggregated,
    R_seasonal_aggregated,
    H_seasonal_aggregated
)

# Extract the updated mean values from the combined state vector
mu_post_SST_D47_seasonal_aggregated = mu_post_seasonal_aggregated[:len(mu_prior_SST_D47_seasonal)]
mu_post_SAT_D47_seasonal_aggregated = mu_post_seasonal_aggregated[len(mu_prior_SST_D47_seasonal):2*len(mu_prior_SST_D47_seasonal)]
mu_post_d18Oc_seasonal_aggregated = mu_post_seasonal_aggregated[2*len(mu_prior_SST_D47_seasonal):3*len(mu_prior_SST_D47_seasonal)]
mu_post_precip_seasonal_aggregated = mu_post_seasonal_aggregated[3*len(mu_prior_d18Oc_seasonal):]

# Extract the updated covariance matrices from the combined covariance matrix
cov_post_SST_D47_seasonal_aggregated = cov_post_seasonal_aggregated[:len(mu_prior_SST_D47_seasonal), :len(mu_prior_SST_D47_seasonal)]
cov_post_SAT_D47_seasonal_aggregated = cov_post_seasonal_aggregated[len(mu_prior_SST_D47_seasonal):2*len(mu_prior_SST_D47_seasonal), len(mu_prior_SST_D47_seasonal):2*len(mu_prior_SST_D47_seasonal)]
cov_post_d18Oc_seasonal_aggregated = cov_post_seasonal_aggregated[2*len(mu_prior_SST_D47_seasonal):3*len(mu_prior_SST_D47_seasonal), 2*len(mu_prior_SST_D47_seasonal):3*len(mu_prior_SST_D47_seasonal)]
cov_post_precip_seasonal_aggregated = cov_post_seasonal_aggregated[3*len(mu_prior_d18Oc_seasonal):, 3*len(mu_prior_d18Oc_seasonal):]

for measurement in Lutetian_seasonally_aggregated_data_dict: # Loop over measurements    
    # Track and update likelihood statistics
    weighted_sum_seasonal_aggregated, effective_weights_total_seasonal_aggregated, n_update_seasonal_aggregated, seasonal_aggregated_data = likelihood_statistics_multi(
        weighted_sum_seasonal_aggregated,
        effective_weights_total_seasonal_aggregated,
        n_update_seasonal_aggregated,
        seasonal_aggregated_data,
        measurement,
        timestamp = "season_score",
        timestamp_sd = "Season_err",
        Variable_names = var_names,
        Variable_names_SDs = var_SD_names
    )


# Normalize the weighted_sum_seasonal to obtain weighted mean
# Calculate inverse square root of the effective_weights_total_seasonal to contain the weighted standard deviation
# Print likelihood statistics
print("Likelihood statistics:")
num_vars = len(var_names)  # number of variables (e.g., D47, d18O)
num_bins_seasonal = int(len(weighted_sum_seasonal_aggregated) / num_vars)

for var_idx, var_name in enumerate(var_names):
    print(f"Results for variable: {var_name}")
    for bin_idx in range(num_bins_seasonal):
        idx = var_idx * num_bins_seasonal + bin_idx
        if effective_weights_total_seasonal_aggregated[idx] is not None and effective_weights_total_seasonal_aggregated[idx] != 0:
            mu_likelihood_seasonal_aggregated[idx] = weighted_sum_seasonal_aggregated[idx] / effective_weights_total_seasonal_aggregated[idx]
            std_likelihood_seasonal_aggregated[idx] = np.sqrt(1 / effective_weights_total_seasonal_aggregated[idx])
        else:
            # If there are no data points for this bin, set the likelihood to NaN
            mu_likelihood_seasonal_aggregated[idx] = np.nan
            std_likelihood_seasonal_aggregated[idx] = np.nan
        print(f"  Bin {bin_idx + 1}:")
        print(f"    Weighted Average: {mu_likelihood_seasonal_aggregated[idx]}")
        print(f"    Aggregated Uncertainty: {std_likelihood_seasonal_aggregated[idx]}")
        print(f"    Number of Data Points: {n_update_seasonal_aggregated[idx]}")
    print()

print("Original Prior Mean SST-D47 Seasonal:\n", mu_prior_SST_D47_seasonal_original)
print("Original Prior Standard Deviation SST-D47 Seasonal:\n", np.sqrt(np.diag(cov_prior_SST_D47_seasonal_original)))
print("Updated Posterior Mean SST-D47 Seasonal:\n", mu_post_SST_D47_seasonal_aggregated)
print("Updated Posterior Standard Deviation SST-D47 Seasonal:\n", np.sqrt(np.diag(cov_post_SST_D47_seasonal_aggregated)))
print("Original Prior Mean SAT-D47 Seasonal:\n", mu_prior_SAT_seasonal_original)
print("Original Prior Standard Deviation SAT-D47 Seasonal:\n", np.sqrt(np.diag(cov_prior_SAT_seasonal_original)))
print("Updated Posterior Mean SAT-D47 Seasonal:\n", mu_post_SAT_D47_seasonal_aggregated)
print("Updated Posterior Standard Deviation SAT-D47 Seasonal:\n", np.sqrt(np.diag(cov_post_SAT_D47_seasonal_aggregated)))
print("Original Prior Mean d18Oc Seasonal:\n", mu_prior_d18Oc_seasonal_original)
print("Original Prior Standard Deviation d18Oc Seasonal:\n", np.sqrt(np.diag(cov_prior_d18Oc_seasonal_original)))
print("Updated Posterior Mean d18Oc Seasonal:\n", mu_post_d18Oc_seasonal_aggregated)
print("Updated Posterior Standard Deviation d18Oc Seasonal:\n", np.sqrt(np.diag(cov_post_d18Oc_seasonal_aggregated)))
print("Original Prior Mean precipitation Seasonal:\n", mu_prior_precip_seasonal_original)
print("Original Prior Standard Deviation precipitation Seasonal:\n", np.sqrt(np.diag(cov_prior_precip_seasonal_original)))
print("Updated Posterior Mean precipitation Seasonal:\n", mu_post_precip_seasonal_aggregated)
print("Updated Posterior Standard Deviation precipitation Seasonal:\n", np.sqrt(np.diag(cov_post_precip_seasonal_aggregated)))

Likelihood statistics:
Results for variable: D47_mean
  Bin 1:
    Weighted Average: 0.599910483
    Aggregated Uncertainty: 0.034247571
    Number of Data Points: 1.0
  Bin 2:
    Weighted Average: 0.574933608
    Aggregated Uncertainty: 0.032719573
    Number of Data Points: 1.0
  Bin 3:
    Weighted Average: 0.589860492
    Aggregated Uncertainty: 0.042653301
    Number of Data Points: 1.0
  Bin 4:
    Weighted Average: 0.58302926
    Aggregated Uncertainty: 0.030150341
    Number of Data Points: 1.0

Results for variable: d18O
  Bin 1:
    Weighted Average: -0.159068042
    Aggregated Uncertainty: 0.238708078
    Number of Data Points: 1.0
  Bin 2:
    Weighted Average: -1.531684703
    Aggregated Uncertainty: 0.212594892
    Number of Data Points: 1.0
  Bin 3:
    Weighted Average: -2.150527044
    Aggregated Uncertainty: 0.312787802
    Number of Data Points: 1.0
  Bin 4:
    Weighted Average: -1.589938113
    Aggregated Uncertainty: 0.188987236
    Number of Data Points: 1.0

Original Prior Mean SST-D47 Seasonal:
 [0.59558454 0.5944782  0.5693298  0.57532305]
Original Prior Standard Deviation SST-D47 Seasonal:
 [0.00921641 0.00955641 0.00823186 0.00872053]
Updated Posterior Mean SST-D47 Seasonal:
 [0.59039957 0.58851496 0.56608958 0.57803973]
Updated Posterior Standard Deviation SST-D47 Seasonal:
 [0.00281963 0.0026522  0.0025446  0.00278496]
Original Prior Mean SAT-D47 Seasonal:
 [16.82159993 22.44667847 34.54556681 26.06112386]
Original Prior Standard Deviation SAT-D47 Seasonal:
 [3.40758273 4.36702297 3.22740902 5.65487419]
Updated Posterior Mean SAT-D47 Seasonal:
 [0.60546524 0.58546743 0.56052517 0.58796684]
Updated Posterior Standard Deviation SAT-D47 Seasonal:
 [0.00959118 0.00828959 0.00770285 0.0091893 ]
Original Prior Mean d18Oc Seasonal:
 [-0.94610631 -1.04113704 -3.08652898 -2.58441643]
Original Prior Standard Deviation d18Oc Seasonal:
 [0.48936498 0.526327   0.65431916 0.79321889]
Updated Posterior Mean d18Oc Seasonal:
 [-0.86608122 -1.01663109 -2.87292332 -1.86973968]
Updated Posterior Standard Deviation d18Oc Seasonal:
 [0.1408767  0.12247934 0.19625317 0.15895546]
Original Prior Mean precipitation Seasonal:
 [0.20279093 0.18348966 0.19218344 0.18032   ]
Original Prior Standard Deviation precipitation Seasonal:
 [0.06518114 0.07164279 0.10863477 0.06570094]
Updated Posterior Mean precipitation Seasonal:
 [0.17888102 0.1149692  0.17635882 0.20151172]
Updated Posterior Standard Deviation precipitation Seasonal:
 [0.06333679 0.06099114 0.08593415 0.06465625]

# --- D47 ---

# Plot the updated seasonal posterior for SST_D47
std_post_SST_D47_seasonal_aggregated = np.sqrt(np.diag(cov_post_SST_D47_seasonal_aggregated))
std_prior_SST_D47_seasonal_original = np.sqrt(np.diag(cov_prior_SST_D47_seasonal_original))
var_start_D47_seasonal = var_names.index("D47_mean") * num_bins_seasonal # Determine the start index for the D47 variable
var_end_D47_seasonal = var_start_D47_seasonal + num_bins_seasonal # Determine the end index for the D47 variable
n_update_seasonal_aggregated_D47 = n_update_seasonal_aggregated[var_start_D47_seasonal:var_end_D47_seasonal]  # Extract the number of updates for D47

plt.figure(figsize=(10, 6))

# PRIOR
plt.plot(seasons_scale, mu_prior_SST_D47_seasonal_original, label='Prior Mean (CESM model)', color='b', marker='o')
plt.fill_between(
    seasons_scale,
    mu_prior_SST_D47_seasonal_original - stats.t.ppf(1 - 0.025, n_models_seasonal) * std_prior_SST_D47_seasonal_original / np.sqrt(n_models_seasonal),
    mu_prior_SST_D47_seasonal_original + stats.t.ppf(1 - 0.025, n_models_seasonal) * std_prior_SST_D47_seasonal_original / np.sqrt(n_models_seasonal),
    color='b',
    alpha=0.2,
    label='95% Confidence Interval'
)

# LIKELIHOOD
# Determine the start and end indices for the selected variable to parse information from the likelihood statistics
plt.plot(seasons_scale, mu_likelihood_seasonal_aggregated[var_start_D47_seasonal:var_end_D47_seasonal], label='Likelihood Mean (clumped data)', color='y', marker='o')
plt.fill_between(
    seasons_scale,
    mu_likelihood_seasonal_aggregated[var_start_D47_seasonal:var_end_D47_seasonal] - stats.t.ppf(1 - 0.025, n_update_seasonal_aggregated_D47) * std_likelihood_seasonal_aggregated[var_start_D47_seasonal:var_end_D47_seasonal] / np.sqrt(n_update_seasonal_aggregated_D47),
    mu_likelihood_seasonal_aggregated[var_start_D47_seasonal:var_end_D47_seasonal] + stats.t.ppf(1 - 0.025, n_update_seasonal_aggregated_D47) * std_likelihood_seasonal_aggregated[var_start_D47_seasonal:var_end_D47_seasonal] / np.sqrt(n_update_seasonal_aggregated_D47),
    color='y',
    alpha=0.2,
    label='95% Confidence Interval'
)
for measurement in Lutetian_seasonally_aggregated_data_dict:
    plt.plot(measurement["season_score"] + 1, measurement["D47_mean"], color="y", marker="o", alpha=0.2)
    plt.errorbar(measurement["season_score"] + 1, measurement["D47_mean"], yerr=measurement["D47_SD"], color="y", alpha=0.2, capsize=5)
secax = plt.gca().secondary_xaxis('top')
secax.set_xticks(seasons_scale)
secax.set_xticklabels([f"n = {int(n)}" for n in n_update_seasonal_aggregated_D47])
secax.tick_params(axis='x', rotation=0)

# POSTERIOR
plt.plot(seasons_scale, mu_post_SST_D47_seasonal_aggregated, label='Posterior Mean (CESM model + clumped data)', color='r', marker='o')
plt.fill_between(
    seasons_scale,
    mu_post_SST_D47_seasonal_aggregated - stats.t.ppf(1 - 0.025, (n_update_seasonal_aggregated_D47 + n_models_seasonal)) * std_post_SST_D47_seasonal_aggregated / np.sqrt(n_update_seasonal_aggregated_D47 + n_models_seasonal),
    mu_post_SST_D47_seasonal_aggregated + stats.t.ppf(1 - 0.025, (n_update_seasonal_aggregated_D47 + n_models_seasonal)) * std_post_SST_D47_seasonal_aggregated / np.sqrt(n_update_seasonal_aggregated_D47 + n_models_seasonal),
    color='r',
    alpha=0.2,
    label='95% Confidence Interval (Posterior)'
)

plt.xticks(seasons_scale, seasons)
plt.title('Prior and posterior Mean and 95% Confidence Interval for Seasonal SST D47 values\n(Based on seasonal averages per specimen)')
plt.xlabel('Season')
plt.ylabel('SST D47 value')
plt.legend()
plt.grid(True)
plt.show()

# ---d18Oc---

# Plot the updated seasonal posterior for d18Oc
std_post_d18Oc_seasonal_aggregated = np.sqrt(np.diag(cov_post_d18Oc_seasonal_aggregated))
std_prior_d18Oc_seasonal_original = np.sqrt(np.diag(cov_prior_d18Oc_seasonal_original))
var_start_d18Oc_seasonal = var_names.index("d18O") * num_bins_seasonal # Determine the start index for the d18Oc variable
var_end_d18Oc_seasonal = var_start_d18Oc_seasonal + num_bins_seasonal # Determine the end index for the d18Oc variable
n_update_seasonal_d18Oc = n_update_seasonal_aggregated[var_start_d18Oc_seasonal:var_end_d18Oc_seasonal]  # Extract the number of updates for d18Oc

plt.figure(figsize=(10, 6))

# PRIOR
plt.plot(seasons_scale, mu_prior_d18Oc_seasonal_original, label='Prior Mean (CESM model)', color='b', marker='o')
plt.fill_between(
    seasons_scale,
    mu_prior_d18Oc_seasonal_original - stats.t.ppf(1 - 0.025, n_models_seasonal) * std_prior_d18Oc_seasonal_original / np.sqrt(n_models_seasonal),
    mu_prior_d18Oc_seasonal_original + stats.t.ppf(1 - 0.025, n_models_seasonal) * std_prior_d18Oc_seasonal_original / np.sqrt(n_models_seasonal),
    color='b',
    alpha=0.2,
    label='95% Confidence Interval'
)

# LIKELIHOOD
# Determine the start and end indices for the selected variable
plt.plot(seasons_scale, mu_likelihood_seasonal_aggregated[var_start_d18Oc_seasonal:var_end_d18Oc_seasonal], label='Likelihood Mean (clumped data)', color='y', marker='o')
plt.fill_between(
    seasons_scale,
    mu_likelihood_seasonal_aggregated[var_start_d18Oc_seasonal:var_end_d18Oc_seasonal] - stats.t.ppf(1 - 0.025, n_update_seasonal_d18Oc) * std_likelihood_seasonal_aggregated[var_start_d18Oc_seasonal:var_end_d18Oc_seasonal] / np.sqrt(n_update_seasonal_d18Oc),
    mu_likelihood_seasonal_aggregated[var_start_d18Oc_seasonal:var_end_d18Oc_seasonal] + stats.t.ppf(1 - 0.025, n_update_seasonal_d18Oc) * std_likelihood_seasonal_aggregated[var_start_d18Oc_seasonal:var_end_d18Oc_seasonal] / np.sqrt(n_update_seasonal_d18Oc),
    color='y',
    alpha=0.2,
    label='95% Confidence Interval'
)
for measurement in Lutetian_seasonally_aggregated_data_dict:
    plt.plot(measurement["season_score"] + 1, measurement["d18O"], color="y", marker="o", alpha=0.2)
    plt.errorbar(measurement["season_score"] + 1, measurement["d18O"], yerr=measurement["d18O_SD"], color="y", alpha=0.2, capsize=5)
secax = plt.gca().secondary_xaxis('top')
secax.set_xticks(seasons_scale)
secax.set_xticklabels([f"n = {int(n)}" for n in n_update_seasonal_d18Oc])
secax.tick_params(axis='x', rotation=0)

# POSTERIOR
plt.plot(seasons_scale, mu_post_d18Oc_seasonal_aggregated, label='Posterior Mean (CESM model + clumped data)', color='r', marker='o')
plt.fill_between(
    seasons_scale,
    mu_post_d18Oc_seasonal_aggregated - stats.t.ppf(1 - 0.025, (n_update_seasonal_d18Oc + n_models_seasonal)) * std_post_d18Oc_seasonal_aggregated / np.sqrt(n_update_seasonal_d18Oc + n_models_seasonal),
    mu_post_d18Oc_seasonal_aggregated + stats.t.ppf(1 - 0.025, (n_update_seasonal_d18Oc + n_models_seasonal)) * std_post_d18Oc_seasonal_aggregated / np.sqrt(n_update_seasonal_d18Oc + n_models_seasonal),
    color='r',
    alpha=0.2,
    label='95% Confidence Interval (Posterior)'
)

plt.xticks(seasons_scale, seasons)
plt.title('Prior and posterior Mean and 95% Confidence Interval for Seasonal d18Oc values\n(Based on seasonal averages per specimen)')
plt.xlabel('Season')
plt.ylabel('d18Oc value')
plt.legend()
plt.grid(True)
plt.show()

# Apply Kalman function to update the prior with monthly data including updating the prec estimates
# Update the monthly D47 and prec prior with all measurements using block updating
monthly_aggregated_data = {} # Keep track of datapoints per season
n_update_monthly_aggregated = np.concatenate([mu_prior_SST_D47_monthly * 0, mu_prior_d18Oc_monthly * 0]) # Vector to store sample size per season for confidence interval plotting
weighted_sum_monthly_aggregated = np.concatenate([mu_prior_SST_D47_monthly * 0, mu_prior_d18Oc_monthly * 0]) # Vector to store mean temperature per season for confidence interval plotting
effective_weights_total_monthly_aggregated = np.concatenate([mu_prior_SST_D47_monthly * 0, mu_prior_d18Oc_monthly * 0]) # Vector to store temperature uncertainty per season for confidence interval plotting
mu_likelihood_monthly_aggregated = np.concatenate([mu_prior_SST_D47_monthly * 0, mu_prior_d18Oc_monthly * 0]) # Vector to store mean temperature per season for confidence interval plotting
std_likelihood_monthly_aggregated = np.concatenate([mu_prior_SST_D47_monthly * 0, mu_prior_d18Oc_monthly * 0]) # Vector to store temperature uncertainty per season for confidence interval plotting
var_names = ["D47_mean", "d18O"] # List of variable names which are updated
var_SD_names = ["D47_SD", "d18O_SD"] # List of names of variable uncertainties which are updated

# Update the prior with monthly data using the Kalman filter in block updating form
mu_post_monthly_aggregated, cov_post_monthly_aggregated = kalman_update_block(
    mu_prior_monthly_combined,
    cov_prior_monthly_combined,
    Z_monthly_aggregated,
    R_monthly_aggregated,
    H_monthly_aggregated
)

# Extract the updated mean values from the combined state vector
mu_post_SST_D47_monthly_aggregated = mu_post_monthly_aggregated[:len(mu_prior_SST_D47_monthly)]
mu_post_SAT_D47_monthly_aggregated = mu_post_monthly_aggregated[len(mu_prior_SST_D47_monthly):2*len(mu_prior_SST_D47_monthly)]
mu_post_d18Oc_monthly_aggregated = mu_post_monthly_aggregated[2*len(mu_prior_SST_D47_monthly):3*len(mu_prior_SST_D47_monthly)]
mu_post_precip_monthly_aggregated = mu_post_monthly_aggregated[3*len(mu_prior_d18Oc_monthly):]

# Extract the updated covariance matrices from the combined covariance matrix
cov_post_SST_D47_monthly_aggregated = cov_post_monthly_aggregated[:len(mu_prior_SST_D47_monthly), :len(mu_prior_SST_D47_monthly)]
cov_post_SAT_D47_monthly_aggregated = cov_post_monthly_aggregated[len(mu_prior_SST_D47_monthly):2*len(mu_prior_SST_D47_monthly), len(mu_prior_SST_D47_monthly):2*len(mu_prior_SST_D47_monthly)]
cov_post_d18Oc_monthly_aggregated = cov_post_monthly_aggregated[2*len(mu_prior_SST_D47_monthly):3*len(mu_prior_SST_D47_monthly), 2*len(mu_prior_SST_D47_monthly):3*len(mu_prior_SST_D47_monthly)]
cov_post_precip_monthly_aggregated = cov_post_monthly_aggregated[3*len(mu_prior_d18Oc_monthly):, 3*len(mu_prior_d18Oc_monthly):]

for measurement in Lutetian_monthly_aggregated_data_dict: # Loop over measurements    
    # Track and update likelihood statistics
    weighted_sum_monthly_aggregated, effective_weights_total_monthly_aggregated, n_update_monthly_aggregated, monthly_aggregated_data = likelihood_statistics_multi(
        weighted_sum_monthly_aggregated,
        effective_weights_total_monthly_aggregated,
        n_update_monthly_aggregated,
        monthly_aggregated_data,
        measurement,
        timestamp = "month_score",
        timestamp_sd = "Month_err",
        Variable_names = var_names,
        Variable_names_SDs = var_SD_names
    )

# Normalize the weighted_sum_monthly to obtain weighted mean
# Calculate inverse square root of the effective_weights_total_monthly to contain the weighted standard deviation
# Print likelihood statistics
print("Likelihood statistics:")
num_vars = len(var_names)  # number of variables (e.g., D47, d18O)
num_bins_monthly = int(len(weighted_sum_monthly_aggregated) / num_vars)

for var_idx, var_name in enumerate(var_names):
    print(f"Results for variable: {var_name}")
    for bin_idx in range(num_bins_monthly):
        idx = var_idx * num_bins_monthly + bin_idx
        if effective_weights_total_monthly_aggregated[idx] is not None and effective_weights_total_monthly_aggregated[idx] != 0:
            mu_likelihood_monthly_aggregated[idx] = weighted_sum_monthly_aggregated[idx] / effective_weights_total_monthly_aggregated[idx]
            std_likelihood_monthly_aggregated[idx] = np.sqrt(1 / effective_weights_total_monthly_aggregated[idx])
        else:
            # If there are no data points for this bin, set the likelihood to NaN
            mu_likelihood_monthly_aggregated[idx] = np.nan
            std_likelihood_monthly_aggregated[idx] = np.nan
        print(f"  Bin {bin_idx + 1}:")
        print(f"    Weighted Average: {mu_likelihood_monthly_aggregated[idx]}")
        print(f"    Aggregated Uncertainty: {std_likelihood_monthly_aggregated[idx]}")
        print(f"    Number of Data Points: {n_update_monthly_aggregated[idx]}")
    print()

print("Original Prior Mean SST-D47 monthly:\n", mu_prior_SST_D47_monthly_original)
print("Original Prior Standard Deviation SST-D47 monthly:\n", np.sqrt(np.diag(cov_prior_SST_D47_monthly_original)))
print("Updated Posterior Mean SST-D47 monthly:\n", mu_post_SST_D47_monthly_aggregated)
print("Updated Posterior Standard Deviation SST-D47 monthly:\n", np.sqrt(np.diag(cov_post_SST_D47_monthly_aggregated)))
print("Original Prior Mean SAT-D47 monthly:\n", mu_prior_SAT_monthly_original)
print("Original Prior Standard Deviation SAT-D47 monthly:\n", np.sqrt(np.diag(cov_prior_SAT_monthly_original)))
print("Updated Posterior Mean SAT-D47 monthly:\n", mu_post_SAT_D47_monthly_aggregated)
print("Updated Posterior Standard Deviation SAT-D47 monthly:\n", np.sqrt(np.diag(cov_post_SAT_D47_monthly_aggregated)))
print("Original Prior Mean d18Oc monthly:\n", mu_prior_d18Oc_monthly_original)
print("Original Prior Standard Deviation d18Oc monthly:\n", np.sqrt(np.diag(cov_prior_d18Oc_monthly_original)))
print("Updated Posterior Mean d18Oc monthly:\n", mu_post_d18Oc_monthly_aggregated)
print("Updated Posterior Standard Deviation d18Oc monthly:\n", np.sqrt(np.diag(cov_post_d18Oc_monthly_aggregated)))
print("Original Prior Mean precipitation monthly:\n", mu_prior_precip_monthly_original)
print("Original Prior Standard Deviation precipitation monthly:\n", np.sqrt(np.diag(cov_prior_precip_monthly_original)))
print("Updated Posterior Mean precipitation monthly:\n", mu_post_precip_monthly_aggregated)
print("Updated Posterior Standard Deviation precipitation monthly:\n", np.sqrt(np.diag(cov_post_precip_monthly_aggregated)))

Likelihood statistics:
Results for variable: D47_mean
  Bin 1:
    Weighted Average: 0.6315000000000001
    Aggregated Uncertainty: 0.0145
    Number of Data Points: 1.0
  Bin 2:
    Weighted Average: 0.5964285714285714
    Aggregated Uncertainty: 0.01096096971726759
    Number of Data Points: 1.0
  Bin 3:
    Weighted Average: 0.6065
    Aggregated Uncertainty: 0.010253048327204941
    Number of Data Points: 1.0
  Bin 4:
    Weighted Average: nan
    Aggregated Uncertainty: nan
    Number of Data Points: 0.0
  Bin 5:
    Weighted Average: 0.5749687499999999
    Aggregated Uncertainty: 0.0051265241636024705
    Number of Data Points: 1.0
  Bin 6:
    Weighted Average: 0.5925625
    Aggregated Uncertainty: 0.00725
    Number of Data Points: 1.0
  Bin 7:
    Weighted Average: 0.5888260869565218
    Aggregated Uncertainty: 0.004275816728492018
    Number of Data Points: 1.0
  Bin 8:
    Weighted Average: nan
    Aggregated Uncertainty: nan
    Number of Data Points: 0.0
  Bin 9:
    Weighted Average: 0.5830000000000001
    Aggregated Uncertainty: 0.004973458969132757
    Number of Data Points: 1.0
  Bin 10:
    Weighted Average: nan
    Aggregated Uncertainty: nan
    Number of Data Points: 0.0
  Bin 11:
    Weighted Average: 0.594
    Aggregated Uncertainty: 0.011839200423452026
    Number of Data Points: 1.0
  Bin 12:
    Weighted Average: 0.5921333333333333
    Aggregated Uncertainty: 0.007487767802667672
    Number of Data Points: 1.0

Results for variable: d18O
  Bin 1:
    Weighted Average: 0.016000000000000014
    Aggregated Uncertainty: 0.0447213595499958
    Number of Data Points: 1.0
  Bin 2:
    Weighted Average: 0.10700000000000001
    Aggregated Uncertainty: 0.0316227766016838
    Number of Data Points: 1.0
  Bin 3:
    Weighted Average: -0.22615384615384615
    Aggregated Uncertainty: 0.02773500981126146
    Number of Data Points: 1.0
  Bin 4:
    Weighted Average: -0.5469999999999999
    Aggregated Uncertainty: 0.01414213562373095
    Number of Data Points: 1.0
  Bin 5:
    Weighted Average: -1.2017647058823526
    Aggregated Uncertainty: 0.009166984970282115
    Number of Data Points: 1.0
  Bin 6:
    Weighted Average: -1.6498290598290601
    Aggregated Uncertainty: 0.009245003270420488
    Number of Data Points: 1.0
  Bin 7:
    Weighted Average: -2.0610810810810816
    Aggregated Uncertainty: 0.009491579957524992
    Number of Data Points: 1.0
  Bin 8:
    Weighted Average: -1.7621904761904763
    Aggregated Uncertainty: 0.009759000729485335
    Number of Data Points: 1.0
  Bin 9:
    Weighted Average: -1.3562711864406782
    Aggregated Uncertainty: 0.009205746178983237
    Number of Data Points: 1.0
  Bin 10:
    Weighted Average: -0.5620833333333334
    Aggregated Uncertainty: 0.014433756729740647
    Number of Data Points: 1.0
  Bin 11:
    Weighted Average: -0.06419354838709676
    Aggregated Uncertainty: 0.017960530202677495
    Number of Data Points: 1.0
  Bin 12:
    Weighted Average: -0.162
    Aggregated Uncertainty: 0.0223606797749979
    Number of Data Points: 1.0

Original Prior Mean SST-D47 monthly:
 [0.59656789 0.59858139 0.59867924 0.59638204 0.58858685 0.57705811
 0.56727922 0.56351195 0.5670409  0.57510508 0.58376364 0.59171077]
Original Prior Standard Deviation SST-D47 monthly:
 [0.00890411 0.00944467 0.00936157 0.00859554 0.00739686 0.00653869
 0.00575527 0.00534817 0.00493985 0.00482531 0.00621776 0.00782822]
Updated Posterior Mean SST-D47 monthly:
 [0.59746433 0.59942274 0.59953844 0.59638204 0.5892276  0.57770401
 0.56711071 0.56351195 0.56748139 0.57510508 0.58307085 0.5924027 ]
Updated Posterior Standard Deviation SST-D47 monthly:
 [1.32467442e-03 8.57442039e-04 5.08756743e-04 7.08126758e-10
 9.15449230e-04 8.96916623e-04 3.84829259e-04 5.20625146e-10
 7.72355884e-04 1.02815228e-09 1.25655914e-03 1.58388923e-03]
Original Prior Mean SAT-D47 monthly:
 [16.27954224 17.01156006 18.66868896 21.54889648 27.12244995 32.0811853
 35.76927734 35.78623779 32.02878784 25.72896606 20.42561768 17.17369751]
Original Prior Standard Deviation SAT-D47 monthly:
 [3.46377896 3.10594374 2.78918732 2.54271169 2.46488387 2.73331005
 2.85535749 2.646315   2.52596201 3.00791784 3.58824735 3.69933077]
Updated Posterior Mean SAT-D47 monthly:
 [0.58384115 0.57819316 0.57212356 0.56538091 0.55283988 0.5417955
 0.5352128  0.53646114 0.54908711 0.56900319 0.58019837 0.58547234]
Updated Posterior Standard Deviation SAT-D47 monthly:
 [0.00205032 0.00212821 0.00216467 0.00213557 0.00197392 0.00179966
 0.00174475 0.00179853 0.0016947  0.00147584 0.00164494 0.00189799]
Original Prior Mean d18Oc monthly:
 [-0.87134431 -0.71613404 -0.7139883  -0.89894578 -1.51047705 -2.43504627
 -3.25136303 -3.57317763 -3.27234184 -2.5945588  -1.88634864 -1.25084057]
Original Prior Standard Deviation d18Oc monthly:
 [0.43074162 0.4079074  0.39792762 0.39889866 0.41067086 0.41700845
 0.43048772 0.48872905 0.56171093 0.58936283 0.51749387 0.46825537]
Updated Posterior Mean d18Oc monthly:
 [-0.0594466  -0.16160171 -0.22093518 -0.23611568 -0.55835238 -1.31067957
 -1.89449543 -2.11835829 -1.58325457 -0.69231214 -0.34526637 -0.09119921]
Updated Posterior Standard Deviation d18Oc monthly:
 [0.03807578 0.03552346 0.03066903 0.02830315 0.02933908 0.03518046
 0.04691073 0.04543986 0.04263553 0.05642311 0.04493518 0.03952293]
Original Prior Mean precipitation monthly:
 [0.20503649 0.20900211 0.21310989 0.19671564 0.14064345 0.16901619
 0.20951558 0.19801853 0.19421264 0.16398235 0.18276501 0.19433419]
Original Prior Standard Deviation precipitation monthly:
 [0.06656403 0.06261988 0.05604318 0.06767304 0.07139255 0.09812866
 0.12533398 0.10071089 0.0756273  0.0488595  0.0686841  0.0680282 ]
Updated Posterior Mean precipitation monthly:
 [0.10559622 0.12571749 0.17342212 0.21719024 0.16255251 0.19073496
 0.2447648  0.21186652 0.21632807 0.22008681 0.15111996 0.12621666]
Updated Posterior Standard Deviation precipitation monthly:
 [0.04475803 0.04302467 0.03229094 0.03067075 0.02205757 0.03134652
 0.04153917 0.04164468 0.05090759 0.03852647 0.02911428 0.02952848]

# --- D47 ---

# Plot the updated monthly posterior for SST_D47
std_post_SST_D47_monthly_aggregated = np.sqrt(np.diag(cov_post_SST_D47_monthly_aggregated))
std_prior_SST_D47_monthly_original = np.sqrt(np.diag(cov_prior_SST_D47_monthly_original))
var_start_D47_monthly = var_names.index("D47_mean") * num_bins_monthly # Determine the start index for the D47 variable
var_end_D47_monthly = var_start_D47_monthly + num_bins_monthly # Determine the end index for the D47 variable
n_update_monthly_aggregated_D47 = n_update_monthly_aggregated[var_start_D47_monthly:var_end_D47_monthly]  # Extract the number of updates for D47

plt.figure(figsize=(10, 6))

# PRIOR
plt.plot(months_scale, mu_prior_SST_D47_monthly_original, label='Prior Mean (CESM model)', color='b', marker='o')
plt.fill_between(
    months_scale,
    mu_prior_SST_D47_monthly_original - stats.t.ppf(1 - 0.025, n_models_monthly) * std_prior_SST_D47_monthly_original / np.sqrt(n_models_monthly),
    mu_prior_SST_D47_monthly_original + stats.t.ppf(1 - 0.025, n_models_monthly) * std_prior_SST_D47_monthly_original / np.sqrt(n_models_monthly),
    color='b',
    alpha=0.2,
    label='95% Confidence Interval'
)

# LIKELIHOOD
# Determine the start and end indices for the selected variable to parse information from the likelihood statistics
plt.plot(months_scale, mu_likelihood_monthly_aggregated[var_start_D47_monthly:var_end_D47_monthly], label='Likelihood Mean (clumped data)', color='y', marker='o')
plt.fill_between(
    months_scale,
    mu_likelihood_monthly_aggregated[var_start_D47_monthly:var_end_D47_monthly] - stats.t.ppf(1 - 0.025, n_update_monthly_aggregated_D47) * std_likelihood_monthly_aggregated[var_start_D47_monthly:var_end_D47_monthly] / np.sqrt(n_update_monthly_aggregated_D47),
    mu_likelihood_monthly_aggregated[var_start_D47_monthly:var_end_D47_monthly] + stats.t.ppf(1 - 0.025, n_update_monthly_aggregated_D47) * std_likelihood_monthly_aggregated[var_start_D47_monthly:var_end_D47_monthly] / np.sqrt(n_update_monthly_aggregated_D47),
    color='y',
    alpha=0.2,
    label='95% Confidence Interval'
)
for measurement in Lutetian_monthly_aggregated_data_dict:
    plt.plot(measurement["month_score"] + 1, measurement["D47_mean"], color="y", marker="o", alpha=0.2)
    plt.errorbar(measurement["month_score"] + 1, measurement["D47_mean"], yerr=measurement["D47_SD"], color="y", alpha=0.2, capsize=5)
secax = plt.gca().secondary_xaxis('top')
secax.set_xticks(months_scale)
secax.set_xticklabels([f"n = {int(n)}" for n in n_update_monthly_aggregated_D47])
secax.tick_params(axis='x', rotation=0)

# POSTERIOR
plt.plot(months_scale, mu_post_SST_D47_monthly_aggregated, label='Posterior Mean (CESM model + clumped data)', color='r', marker='o')
plt.fill_between(
    months_scale,
    mu_post_SST_D47_monthly_aggregated - stats.t.ppf(1 - 0.025, (n_update_monthly_aggregated_D47 + n_models_monthly)) * std_post_SST_D47_monthly_aggregated / np.sqrt(n_update_monthly_aggregated_D47 + n_models_monthly),
    mu_post_SST_D47_monthly_aggregated + stats.t.ppf(1 - 0.025, (n_update_monthly_aggregated_D47 + n_models_monthly)) * std_post_SST_D47_monthly_aggregated / np.sqrt(n_update_monthly_aggregated_D47 + n_models_monthly),
    color='r',
    alpha=0.2,
    label='95% Confidence Interval (Posterior)'
)

plt.xticks(months_scale, month_names)
plt.title('Prior and posterior Mean and 95% Confidence Interval for monthly SST D47 values\n(Based on monthly averages per specimen)')
plt.xlabel('Season')
plt.ylabel('SST D47 value')
plt.legend()
plt.grid(True)
plt.show()

# ---d18Oc---

# Plot the updated monthly posterior for d18Oc
std_post_d18Oc_monthly_aggregated = np.sqrt(np.diag(cov_post_d18Oc_monthly_aggregated))
std_prior_d18Oc_monthly_original = np.sqrt(np.diag(cov_prior_d18Oc_monthly_original))
var_start_d18Oc_monthly = var_names.index("d18O") * num_bins_monthly # Determine the start index for the d18Oc variable
var_end_d18Oc_monthly = var_start_d18Oc_monthly + num_bins_monthly # Determine the end index for the d18Oc variable
n_update_monthly_aggregated_d18Oc = n_update_monthly_aggregated[var_start_d18Oc_monthly:var_end_d18Oc_monthly]  # Extract the number of updates for d18Oc

plt.figure(figsize=(10, 6))

# PRIOR
plt.plot(months_scale, mu_prior_d18Oc_monthly_original, label='Prior Mean (CESM model)', color='b', marker='o')
plt.fill_between(
    months_scale,
    mu_prior_d18Oc_monthly_original - stats.t.ppf(1 - 0.025, n_models_monthly) * std_prior_d18Oc_monthly_original / np.sqrt(n_models_monthly),
    mu_prior_d18Oc_monthly_original + stats.t.ppf(1 - 0.025, n_models_monthly) * std_prior_d18Oc_monthly_original / np.sqrt(n_models_monthly),
    color='b',
    alpha=0.2,
    label='95% Confidence Interval'
)

# LIKELIHOOD
# Determine the start and end indices for the selected variable
plt.plot(months_scale, mu_likelihood_monthly_aggregated[var_start_d18Oc_monthly:var_end_d18Oc_monthly], label='Likelihood Mean (clumped data)', color='y', marker='o')
plt.fill_between(
    months_scale,
    mu_likelihood_monthly_aggregated[var_start_d18Oc_monthly:var_end_d18Oc_monthly] - stats.t.ppf(1 - 0.025, n_update_monthly_aggregated_d18Oc) * std_likelihood_monthly_aggregated[var_start_d18Oc_monthly:var_end_d18Oc_monthly] / np.sqrt(n_update_monthly_aggregated_d18Oc),
    mu_likelihood_monthly_aggregated[var_start_d18Oc_monthly:var_end_d18Oc_monthly] + stats.t.ppf(1 - 0.025, n_update_monthly_aggregated_d18Oc) * std_likelihood_monthly_aggregated[var_start_d18Oc_monthly:var_end_d18Oc_monthly] / np.sqrt(n_update_monthly_aggregated_d18Oc),
    color='y',
    alpha=0.2,
    label='95% Confidence Interval'
)
for measurement in Lutetian_monthly_aggregated_data_dict:
    plt.plot(measurement["month_score"] + 1, measurement["d18O"], color="y", marker="o", alpha=0.2)
    plt.errorbar(measurement["month_score"] + 1, measurement["d18O"], yerr=measurement["d18O_SD"], color="y", alpha=0.2, capsize=5)
secax = plt.gca().secondary_xaxis('top')
secax.set_xticks(months_scale)
secax.set_xticklabels([f"n = {int(n)}" for n in n_update_monthly_aggregated_d18Oc])
secax.tick_params(axis='x', rotation=0)

# POSTERIOR
plt.plot(months_scale, mu_post_d18Oc_monthly_aggregated, label='Posterior Mean (CESM model + clumped data)', color='r', marker='o')
plt.fill_between(
    months_scale,
    mu_post_d18Oc_monthly_aggregated - stats.t.ppf(1 - 0.025, (n_update_monthly_aggregated_d18Oc + n_models_monthly)) * std_post_d18Oc_monthly_aggregated / np.sqrt(n_update_monthly_aggregated_d18Oc + n_models_monthly),
    mu_post_d18Oc_monthly_aggregated + stats.t.ppf(1 - 0.025, (n_update_monthly_aggregated_d18Oc + n_models_monthly)) * std_post_d18Oc_monthly_aggregated / np.sqrt(n_update_monthly_aggregated_d18Oc + n_models_monthly),
    color='r',
    alpha=0.2,
    label='95% Confidence Interval (Posterior)'
)

plt.xticks(months_scale, month_names)
plt.title('Prior and posterior Mean and 95% Confidence Interval for monthly d18Oc values\n(Based on monthly averages per specimen)')
plt.xlabel('Season')
plt.ylabel('d18Oc value')
plt.legend()
plt.grid(True)
plt.show()

# Set dimensions of data
n_models_monthly = len(Lutetian_models["Cell"])  # Find the total number of models

# Create list of month names
months = ['ja', 'fb', 'mr', 'ar', 'my', 'jn', 'jl', 'ag', 'sp', 'ot', 'nv', 'dc']

# Create a monthly scale for the x-axis
month_names = ['January', 'February', 'March', 'April', 'May', 'June', 'July', 'August', 'September', 'October', 'November', 'December']  # List full month names
months_scale = np.arange(len(months)) + 1  # Create monthly scale

# Create the figure and axes
fig, axes = plt.subplots(1, 3, figsize=(18, 6), sharex=True)

# Panel 1: Plot the prior distribution for SST and SAT
axes[0].plot(months_scale, mu_prior_SAT_monthly, label='Prior SAT Mean', marker='o', color='r')
axes[0].plot(months_scale, mu_prior_SST_monthly, label='Prior SST Mean', marker='o', color='b')

# Add 95% confidence intervals for SAT
axes[0].fill_between(
    months_scale,
    mu_prior_SAT_monthly - stats.t.ppf(1 - 0.025, n_models_monthly) * std_prior_SAT_monthly / np.sqrt(n_models_monthly),
    mu_prior_SAT_monthly + stats.t.ppf(1 - 0.025, n_models_monthly) * std_prior_SAT_monthly / np.sqrt(n_models_monthly),
    alpha=0.2, color='r', label='SAT 95% CI'
)

# Add 95% confidence intervals for SST
axes[0].fill_between(
    months_scale,
    mu_prior_SST_monthly - stats.t.ppf(1 - 0.025, n_models_monthly) * std_prior_SST_monthly / np.sqrt(n_models_monthly),
    mu_prior_SST_monthly + stats.t.ppf(1 - 0.025, n_models_monthly) * std_prior_SST_monthly / np.sqrt(n_models_monthly),
    alpha=0.2, color='b', label='SST 95% CI'
)

# axes[0].set_title('Prior Mean and 95% Confidence Interval for Monthly SST & SAT Values')
axes[0].set_title('Priors and likelihoods for Lutetian case study')
axes[0].set_ylabel('Temperature (°C)')
axes[0].legend()
axes[0].grid(True)

# Panel 2: Plot the prior distribution for SSS and precipitation
axes[1].plot(months_scale, mu_prior_SSS_monthly, label='Prior SSS Mean', marker='o', color='g')
ax2 = axes[1].twinx()  # Create a secondary y-axis for precipitation
ax2.plot(months_scale, mu_prior_precip_monthly, label='Prior Precipitation Mean', marker='o', color='purple')

# Add 95% confidence intervals for SSS
axes[1].fill_between(
    months_scale,
    mu_prior_SSS_monthly - stats.t.ppf(1 - 0.025, n_models_monthly) * std_prior_SSS_monthly / np.sqrt(n_models_monthly),
    mu_prior_SSS_monthly + stats.t.ppf(1 - 0.025, n_models_monthly) * std_prior_SSS_monthly / np.sqrt(n_models_monthly),
    alpha=0.2, color='g', label='SSS 95% CI'
)

# Add 95% confidence intervals for precipitation
ax2.fill_between(
    months_scale,
    mu_prior_precip_monthly - stats.t.ppf(1 - 0.025, n_models_monthly) * std_prior_precip_monthly / np.sqrt(n_models_monthly),
    mu_prior_precip_monthly + stats.t.ppf(1 - 0.025, n_models_monthly) * std_prior_precip_monthly / np.sqrt(n_models_monthly),
    alpha=0.2, color='purple', label='Precipitation 95% CI'
)

axes[1].set_ylabel('SSS (psu)', color='g')
ax2.set_ylabel('Precipitation (mm/day)', color='purple')
# axes[1].set_title('Prior Mean and 95% Confidence Interval for Monthly SSS & Precipitation Values')
axes[1].legend(loc='upper left')
ax2.legend(loc='upper right')
axes[1].grid(True)

# Panel 3: Plot the likelihood distribution for D47 and d18Oc measurements
# Plot individual (non-aggregated) measurements with uncertainties
ax3 = axes[2].twinx()  # secondary y-axis for d18Oc

# Plot D47 likelihood
# Determine the start and end indices for the selected variable to parse information from the likelihood statistics
axes[2].plot(months_scale, mu_likelihood_monthly_aggregated[var_start_D47_monthly:var_end_D47_monthly], marker='o', label='D47 means', color='darkred')
axes[2].fill_between(
    months_scale,
    mu_likelihood_monthly_aggregated[var_start_D47_monthly:var_end_D47_monthly] - 1.96 * np.sqrt(std_likelihood_monthly_aggregated[var_start_D47_monthly:var_end_D47_monthly]),
    mu_likelihood_monthly_aggregated[var_start_D47_monthly:var_end_D47_monthly] + 1.96 * np.sqrt(std_likelihood_monthly_aggregated[var_start_D47_monthly:var_end_D47_monthly]),
    color='darkred',
    alpha=0.2,
    label='D47 95% CI'
)
for measurement in Lutetian_data_dict:
    x_jitter = measurement["month_score"] + 1 + np.random.uniform(-0.2, 0.2)
    axes[2].plot(x_jitter, measurement["D47_final"], color="darkred", marker="o", alpha=0.2)
    axes[2].errorbar(x_jitter, measurement["D47_final"], yerr=measurement["D47_SD"], color="darkred", alpha=0.2)

# Plot d18Oc likelihood
# Determine the start and end indices for the selected variable
ax3.plot(months_scale, mu_likelihood_monthly_aggregated[var_start_d18Oc_monthly:var_end_d18Oc_monthly], marker='o', label='d18Oc means', color='darkblue')
ax3.fill_between(
    months_scale,
    mu_likelihood_monthly_aggregated[var_start_d18Oc_monthly:var_end_d18Oc_monthly] - 1.96 * np.sqrt(std_likelihood_monthly_aggregated[var_start_d18Oc_monthly:var_end_d18Oc_monthly]),
    mu_likelihood_monthly_aggregated[var_start_d18Oc_monthly:var_end_d18Oc_monthly] + 1.96 * np.sqrt(std_likelihood_monthly_aggregated[var_start_d18Oc_monthly:var_end_d18Oc_monthly]),
    color='darkblue',
    alpha=0.2,
    label='d18Oc 95% CI'
)
for measurement in Lutetian_data_dict:
    x_jitter = measurement["month_score"] + 1 + np.random.uniform(-0.2, 0.2)
    ax3.plot(x_jitter, measurement["d18O"], color="darkblue", marker="o", alpha=0.2)
    ax3.errorbar(x_jitter, measurement["d18O"], yerr=measurement["d18O_SD"], color="darkblue", alpha=0.2)

axes[2].set_ylabel('D47 (per mille I-CDES)', color='darkred')
ax3.set_ylabel('d18Oc (per mille VPDB)', color='darkblue')
axes[2].legend(loc='upper left')
ax3.legend(loc='upper right')
axes[2].grid(True)

# # Update the x-axis with month names
axes[0].set_xticks(months_scale)
axes[0].set_xticklabels(month_names, rotation=45, ha="right")
axes[1].set_xticklabels(month_names, rotation=45, ha="right")
axes[2].set_xticklabels(month_names, rotation=45, ha="right")

# Set tight layout
plt.tight_layout()

# Function to propagate covariance on d18Ow in case of linear d18Oc-d18Ow-T equation
def propagate_cov_d18Ow_linear(cov_c, cov_T, cov_cT, A):
    """
    Propagate covariance for:
        w = c - (B - T)/A (+ 0.27)
    where A and B are scalars and the term 0.27 is used to convert between SMOW and VSMOW, but only A is relevant for covariance

    Inputs:
        cov_c  : (n,n) covariance matrix of d18Oc
        cov_T  : (n,n) covariance matrix of T
        cov_cT : (n,n) cross-covariance matrix Cov(c,T)
        A      : scalar (slope)

    Returns:
        cov_w : (n,n) covariance matrix of d18Ow
    """

    dwdc = 1.0
    dwdT = 1.0 / A

    cov_w = (
        dwdc**2 * cov_c
        + dwdc * dwdT * (cov_cT + cov_cT.T)
        + dwdT**2 * cov_T
    )

    return cov_w

# Convert prior D47 to temp
# Use OGLS23 equation to convert D47 to temperature
mu_prior_SST_D47_seasonal_T = D47c.OGLS23.T47(D47 = mu_prior_SST_D47_seasonal_original, sD47 = cov_prior_SST_D47_seasonal_original, return_covar = True)[0]
cov_prior_SST_D47_seasonal_T = D47c.OGLS23.T47(D47 = mu_prior_SST_D47_seasonal_original, sD47 = cov_prior_SST_D47_seasonal_original, return_covar = True)[1]

# Convert prior d18Oc and temperature to d18Ow
# Use inverted equation from Grossman and Ku (1986) and corrected for the 0.27 per mil offset (Dettmann et al., 1999)
mu_prior_d18Ow_seasonal_T = mu_prior_d18Oc_seasonal_original - (20.6 - mu_prior_SST_D47_seasonal_T) / 4.34 + 0.27 # Use inverted equation from Grossman and Ku (1986) and corrected for the 0.27 per mil offset (Dettmann et al., 1999)
# Calculate d18Oc-SST cross-covariance matrix
cross_cov_prior_d18Oc_SST_seasonal = np.cov(Lutetian_models_seasonal[SST_D47_columns_seasonal + d18Oc_columns_seasonal].dropna(), rowvar=False)[:len(d18Oc_columns_seasonal), len(d18Oc_columns_seasonal):]
# Propagate covariance to d18Ow
cov_prior_d18Ow_seasonal_T = propagate_cov_d18Ow_linear(
    cov_prior_d18Oc_seasonal_original,
    cov_prior_SST_D47_seasonal_T,
    cross_cov_prior_d18Oc_SST_seasonal,
    4.34
)

# Convert prior d18Ow to SSS
# Use the equation from SSS = (d18Ow + 9.300) / 0.274 (Harwood et al., 2008)
mu_prior_SSS_d18Ow_seasonal_T = (mu_prior_d18Ow_seasonal_T + 9.300) / 0.274
cov_prior_SSS_d18Ow_seasonal_T = cov_prior_d18Ow_seasonal_T / (0.274 ** 2)

# Convert likelihood D47 to temp
mu_likelihood_seasonal_aggregated_T = D47c.OGLS23.T47(D47 = mu_likelihood_seasonal_aggregated[var_start_D47_seasonal:var_end_D47_seasonal], sD47 = std_likelihood_seasonal_aggregated[var_start_D47_seasonal:var_end_D47_seasonal], return_covar = True)[0]
cov_likelihood_seasonal_T = D47c.OGLS23.T47(D47 = mu_likelihood_seasonal_aggregated[var_start_D47_seasonal:var_end_D47_seasonal], sD47 = std_likelihood_seasonal_aggregated[var_start_D47_seasonal:var_end_D47_seasonal], return_covar = True)[1]

# Convert likelihood d18Oc and temperature to d18Ow (curently not implemented in data tracking function)
mu_likelihood_d18Ow_seasonal_T = mu_likelihood_seasonal_aggregated[var_start_d18Oc_seasonal:var_end_d18Oc_seasonal] - (20.6 - mu_likelihood_seasonal_aggregated_T) / 4.34 + 0.27
std_likelihood_d18Ow_seasonal_T = np.sqrt(
    np.diag(
        propagate_cov_d18Ow_linear(
            np.diag(std_likelihood_seasonal_aggregated[var_start_d18Oc_seasonal:var_end_d18Oc_seasonal] ** 2),
            cov_likelihood_seasonal_T,
            cross_cov_prior_d18Oc_SST_seasonal,
            4.34
        )
    )
)

# Convert likelihood d18Ow to SSS (currently not implemented in data tracking function)
mu_likelihood_SSS_d18Ow_seasonal_T = (mu_likelihood_d18Ow_seasonal_T + 9.300) / 0.274
std_likelihood_SSS_d18Ow_seasonal_T = std_likelihood_d18Ow_seasonal_T / (0.274 ** 2)

# Convert posterior D47 to temp
mu_post_SST_D47_seasonal_aggregated_T = D47c.OGLS23.T47(D47 = mu_post_SST_D47_seasonal_aggregated, sD47 = cov_post_SST_D47_seasonal_aggregated, return_covar = True)[0]
cov_post_SST_D47_seasonal_aggregated_T = D47c.OGLS23.T47(D47 = mu_post_SST_D47_seasonal_aggregated, sD47 = cov_post_SST_D47_seasonal_aggregated, return_covar = True)[1]

# Convert posterior d18Oc and temperature to d18Ow
mu_post_d18Ow_seasonal_aggregated_T = mu_post_d18Oc_seasonal_aggregated - (20.6 - mu_post_SST_D47_seasonal_aggregated_T) / 4.34 + 0.27
cov_post_d18Ow_seasonal_aggregated_T = propagate_cov_d18Ow_linear(
    cov_post_d18Oc_seasonal_aggregated,
    cov_post_SST_D47_seasonal_aggregated_T,
    cross_cov_prior_d18Oc_SST_seasonal,
    4.34
)

# Convert posterior d18Ow to SSS
mu_post_SSS_d18Ow_seasonal_aggregated_T = (mu_post_d18Ow_seasonal_aggregated_T + 9.300) / 0.274
cov_post_SSS_d18Ow_seasonal_aggregated_T = cov_post_d18Ow_seasonal_aggregated_T / (0.274 ** 2)

# Plot in temperature domain
std_prior_SST_D47_seasonal_T = np.sqrt(np.diag(cov_prior_SST_D47_seasonal_T))
std_prior_SSS_d18Ow_seasonal_T = np.sqrt(np.diag(cov_prior_SSS_d18Ow_seasonal_T))
std_likelihood_seasonal_aggregated_T = np.sqrt(np.diag(cov_likelihood_seasonal_T))
std_post_SST_D47_seasonal_aggregated_T = np.sqrt(np.diag(cov_post_SST_D47_seasonal_aggregated_T))
std_post_SSS_d18Ow_seasonal_aggregated_T = np.sqrt(np.diag(cov_post_SSS_d18Ow_seasonal_aggregated_T))

# Initiate plot
fig, axs = plt.subplots(2, 1, figsize=(10, 12), sharex=True)

# FIRST PANEL: SST Results
# PRIOR
axs[0].plot(seasons_scale, mu_prior_SST_D47_seasonal_T, label='Prior Mean (PlioMIP models)', color='b', marker='o')
axs[0].fill_between(seasons_scale,
    mu_prior_SST_D47_seasonal_T - stats.t.ppf(1 - 0.025, n_models_seasonal) * std_prior_SST_D47_seasonal_T / np.sqrt(n_models_seasonal),
    mu_prior_SST_D47_seasonal_T + stats.t.ppf(1 - 0.025, n_models_seasonal) * std_prior_SST_D47_seasonal_T / np.sqrt(n_models_seasonal),
    color='b', alpha=0.2, label='95% Confidence Interval')

# LIKELIHOOD
axs[0].plot(seasons_scale, mu_likelihood_seasonal_aggregated_T, label='Likelihood Mean (clumped data)', color='y', marker='o')
axs[0].fill_between(seasons_scale,
    mu_likelihood_seasonal_aggregated_T - stats.t.ppf(1 - 0.025, n_update_seasonal_aggregated_D47) * std_likelihood_seasonal_aggregated_T / np.sqrt(n_update_seasonal_aggregated_D47),
    mu_likelihood_seasonal_aggregated_T + stats.t.ppf(1 - 0.025, n_update_seasonal_aggregated_D47) * std_likelihood_seasonal_aggregated_T / np.sqrt(n_update_seasonal_aggregated_D47),
    color='y', alpha=0.2, label='95% Confidence Interval')

# POSTERIOR
axs[0].plot(seasons_scale, mu_post_SST_D47_seasonal_aggregated_T, label='Posterior Mean (PlioMIP models + clumped data)', color='r', marker='o')
axs[0].fill_between(seasons_scale,
    mu_post_SST_D47_seasonal_aggregated_T - stats.t.ppf(1 - 0.025, (n_update_seasonal_aggregated_D47 + n_models_seasonal)) * std_post_SST_D47_seasonal_aggregated_T / np.sqrt(n_update_seasonal_aggregated_D47 + n_models_seasonal),
    mu_post_SST_D47_seasonal_aggregated_T + stats.t.ppf(1 - 0.025, (n_update_seasonal_aggregated_D47 + n_models_seasonal)) * std_post_SST_D47_seasonal_aggregated_T / np.sqrt(n_update_seasonal_aggregated_D47 + n_models_seasonal),
    color='r', alpha=0.2, label='95% Confidence Interval (Posterior)')

# Layout for SST panel
axs[0].set_title('Seasonal Sea Surface Temperatures (SST)')
axs[0].set_ylabel('Temperature (°C)')
axs[0].legend(loc='upper left')
axs[0].grid(True)

# SECOND PANEL: SSS Results
# PRIOR
axs[1].plot(seasons_scale, mu_prior_SSS_d18Ow_seasonal_T, label='Prior Mean (PlioMIP models)', color='b', marker='o')
axs[1].fill_between(seasons_scale,
    mu_prior_SSS_d18Ow_seasonal_T - stats.t.ppf(1 - 0.025, n_models_seasonal) * std_prior_SSS_d18Ow_seasonal_T / np.sqrt(n_models_seasonal),
    mu_prior_SSS_d18Ow_seasonal_T + stats.t.ppf(1 - 0.025, n_models_seasonal) * std_prior_SSS_d18Ow_seasonal_T / np.sqrt(n_models_seasonal),
    color='b', alpha=0.2, label='95% Confidence Interval')

# LIKELIHOOD
axs[1].plot(seasons_scale, mu_likelihood_SSS_d18Ow_seasonal_T, label='Likelihood Mean (clumped data)', color='y', marker='o')
axs[1].fill_between(seasons_scale,
    mu_likelihood_SSS_d18Ow_seasonal_T - stats.t.ppf(1 - 0.025, n_update_seasonal_d18Oc) * std_likelihood_SSS_d18Ow_seasonal_T / np.sqrt(n_update_seasonal_d18Oc),
    mu_likelihood_SSS_d18Ow_seasonal_T + stats.t.ppf(1 - 0.025, n_update_seasonal_d18Oc) * std_likelihood_SSS_d18Ow_seasonal_T / np.sqrt(n_update_seasonal_d18Oc),
    color='y', alpha=0.2, label='95% Confidence Interval')

# POSTERIOR
axs[1].plot(seasons_scale, mu_post_SSS_d18Ow_seasonal_aggregated_T, label='Posterior Mean (PlioMIP models + clumped data)', color='r', marker='o')
axs[1].fill_between(seasons_scale,
    mu_post_SSS_d18Ow_seasonal_aggregated_T - stats.t.ppf(1 - 0.025, (n_update_seasonal_d18Oc + n_models_seasonal)) * std_post_SSS_d18Ow_seasonal_aggregated_T / np.sqrt(n_update_seasonal_d18Oc + n_models_seasonal),
    mu_post_SSS_d18Ow_seasonal_aggregated_T + stats.t.ppf(1 - 0.025, (n_update_seasonal_d18Oc + n_models_seasonal)) * std_post_SSS_d18Ow_seasonal_aggregated_T / np.sqrt(n_update_seasonal_d18Oc + n_models_seasonal),
    color='r', alpha=0.2, label='95% Confidence Interval (Posterior)')

# Layout for SSS panel
axs[1].set_title('Seasonal Sea Surface Salinity (SSS)')
axs[1].set_xlabel('Season')
axs[1].set_ylabel('Salinity (PSU)')
axs[1].legend(loc='upper left')
axs[1].grid(True)

# Shared x-axis labels
plt.xticks(seasons_scale, seasons, rotation=45, ha="right")
plt.tight_layout()
plt.show()

# Convert posterior D47 to temp
mu_post_SAT_D47_seasonal_aggregated_T = D47c.OGLS23.T47(D47 = mu_post_SAT_D47_seasonal_aggregated, sD47 = cov_post_SAT_D47_seasonal_aggregated, return_covar = True)[0]
cov_post_SAT_D47_seasonal_aggregated_T = D47c.OGLS23.T47(D47 = mu_post_SAT_D47_seasonal_aggregated, sD47 = cov_post_SAT_D47_seasonal_aggregated, return_covar = True)[1]

# Convert posterior SAT-D47 back to temperature
std_post_SAT_D47_seasonal_aggregated_T = np.nan_to_num(np.sqrt(np.diag(cov_post_SAT_D47_seasonal_aggregated_T)))

fig, axs = plt.subplots(2, 1, figsize=(10, 12), sharex=True)

# --- SAT prior and posterior ---
# PRIOR
axs[0].plot(seasons_scale, mu_prior_SAT_seasonal_original, label='Prior Mean (PlioMIP models)', color='b', marker='o')
axs[0].fill_between(
    seasons_scale,
    mu_prior_SAT_seasonal_original - stats.t.ppf(1 - 0.025, n_models_seasonal) * std_prior_SAT_seasonal / np.sqrt(n_models_seasonal),
    mu_prior_SAT_seasonal_original + stats.t.ppf(1 - 0.025, n_models_seasonal) * std_prior_SAT_seasonal / np.sqrt(n_models_seasonal),
    color='b', alpha=0.2, label='95% Confidence Interval'
)

# POSTERIOR
axs[0].plot(seasons_scale, mu_post_SAT_D47_seasonal_aggregated_T, label='Posterior Mean (PlioMIP models + clumped data)', color='r', marker='o')
axs[0].fill_between(
    seasons_scale,
    mu_post_SAT_D47_seasonal_aggregated_T - stats.t.ppf(1 - 0.025, (n_update_seasonal_aggregated_D47 + n_models_seasonal)) * std_post_SAT_D47_seasonal_aggregated_T / np.sqrt(n_update_seasonal_aggregated_D47 + n_models_seasonal),
    mu_post_SAT_D47_seasonal_aggregated_T + stats.t.ppf(1 - 0.025, (n_update_seasonal_aggregated_D47 + n_models_seasonal)) * std_post_SAT_D47_seasonal_aggregated_T / np.sqrt(n_update_seasonal_aggregated_D47 + n_models_seasonal),
    color='r', alpha=0.2, label='95% Confidence Interval (Posterior)'
)

axs[0].set_title('Posterior Mean and 95% Confidence Interval for Seasonal Surface Air Temperatures\n(Based on seasonal averages per specimen)')
axs[0].set_ylabel('Temperature (°C)')
axs[0].set_ylim(0, 40)
axs[0].legend(loc='upper left')
axs[0].grid(True)

# --- Precipitation prior and posterior ---
# PRIOR
axs[1].plot(seasons_scale, mu_prior_precip_seasonal_original, label='Prior Mean (PlioMIP models)', color='b', marker='o')
axs[1].fill_between(
    seasons_scale,
    mu_prior_precip_seasonal_original - stats.t.ppf(1 - 0.025, n_models_seasonal) * std_prior_precip_seasonal / np.sqrt(n_models_seasonal),
    mu_prior_precip_seasonal_original + stats.t.ppf(1 - 0.025, n_models_seasonal) * std_prior_precip_seasonal / np.sqrt(n_models_seasonal),
    color='b', alpha=0.2, label='95% Confidence Interval'
)

# POSTERIOR
axs[1].plot(seasons_scale, mu_post_precip_seasonal_aggregated, label='Posterior Mean (PlioMIP models + clumped data)', color='r', marker='o')
axs[1].fill_between(
    seasons_scale,
    mu_post_precip_seasonal_aggregated - stats.t.ppf(1 - 0.025, (n_update_seasonal_aggregated_D47 + n_models_seasonal)) * np.sqrt(np.diag(cov_post_precip_seasonal_aggregated)) / np.sqrt(n_update_seasonal_aggregated_D47 + n_models_seasonal),
    mu_post_precip_seasonal_aggregated + stats.t.ppf(1 - 0.025, (n_update_seasonal_aggregated_D47 + n_models_seasonal)) * np.sqrt(np.diag(cov_post_precip_seasonal_aggregated)) / np.sqrt(n_update_seasonal_aggregated_D47 + n_models_seasonal),
    color='r', alpha=0.2, label='95% Confidence Interval (Posterior)'
)

axs[1].set_title('Posterior Mean and 95% Confidence Interval for Seasonal Precipitation')
axs[1].set_xlabel('Season')
axs[1].set_ylabel('Precipitation (mm/day)')
axs[1].legend(loc='upper left')
axs[1].grid(True)

plt.xticks(seasons_scale, seasons, rotation=45, ha="right")
plt.tight_layout()
plt.show()

# Create a figure with two rows and two columns of subplots
fig, axes = plt.subplots(2, 2, figsize=(18, 12))

# --- SST & SAT Prior and Posterior ---
# Prior SST & SAT
axes[0, 0].plot(seasons_scale, mu_prior_SST_seasonal, label='Prior SST Mean', marker='o')
axes[0, 0].plot(seasons_scale, mu_prior_SAT_seasonal, label='Prior SAT Mean', marker='o')
axes[0, 0].fill_between(
    seasons_scale,
    mu_prior_SST_seasonal - stats.t.ppf(1 - 0.025, n_models_seasonal) * std_prior_SST_seasonal / np.sqrt(n_models_seasonal),
    mu_prior_SST_seasonal + stats.t.ppf(1 - 0.025, n_models_seasonal) * std_prior_SST_seasonal / np.sqrt(n_models_seasonal),
    alpha=0.2, label='95% Confidence Interval (SST)'
)
axes[0, 0].fill_between(
    seasons_scale,
    mu_prior_SAT_seasonal - stats.t.ppf(1 - 0.025, n_models_seasonal) * std_prior_SAT_seasonal / np.sqrt(n_models_seasonal),
    mu_prior_SAT_seasonal + stats.t.ppf(1 - 0.025, n_models_seasonal) * std_prior_SAT_seasonal / np.sqrt(n_models_seasonal),
    alpha=0.2, label='95% Confidence Interval (SAT)'
)
axes[0, 0].set_xticks(seasons_scale)
axes[0, 0].set_xticklabels(seasons)
axes[0, 0].set_title('Prior Mean and 95% CI for Seasonal SST & SAT')
axes[0, 0].set_xlabel('Season')
axes[0, 0].set_ylabel('Temperature (°C)')
axes[0, 0].set_ylim(0, 45)
axes[0, 0].legend(loc='upper left')
axes[0, 0].grid(True)

# Posterior SST & SAT
axes[0, 1].plot(seasons_scale, mu_post_SST_D47_seasonal_aggregated_T, label='Posterior SST Mean', marker='o')
axes[0, 1].plot(seasons_scale, mu_post_SAT_D47_seasonal_aggregated_T, label='Posterior SAT Mean', marker='o')
axes[0, 1].fill_between(
    seasons_scale,
    mu_post_SST_D47_seasonal_aggregated_T - stats.t.ppf(1 - 0.025, n_update_seasonal_aggregated_D47 + n_models_seasonal) * std_post_SST_D47_seasonal_aggregated_T / np.sqrt(n_update_seasonal_aggregated_D47 + n_models_seasonal),
    mu_post_SST_D47_seasonal_aggregated_T + stats.t.ppf(1 - 0.025, n_update_seasonal_aggregated_D47 + n_models_seasonal) * std_post_SST_D47_seasonal_aggregated_T / np.sqrt(n_update_seasonal_aggregated_D47 + n_models_seasonal),
    alpha=0.2, label='95% Confidence Interval (SST)'
)
axes[0, 1].fill_between(
    seasons_scale,
    mu_post_SAT_D47_seasonal_aggregated_T - stats.t.ppf(1 - 0.025, n_update_seasonal_aggregated_D47 + n_models_seasonal) * std_post_SAT_D47_seasonal_aggregated_T / np.sqrt(n_update_seasonal_aggregated_D47 + n_models_seasonal),
    mu_post_SAT_D47_seasonal_aggregated_T + stats.t.ppf(1 - 0.025, n_update_seasonal_aggregated_D47 + n_models_seasonal) * std_post_SAT_D47_seasonal_aggregated_T / np.sqrt(n_update_seasonal_aggregated_D47 + n_models_seasonal),
    alpha=0.2, label='95% Confidence Interval (SAT)'
)
axes[0, 1].set_xticks(seasons_scale)
axes[0, 1].set_xticklabels(seasons)
axes[0, 1].set_title('Posterior Mean and 95% CI for Seasonal SST & SAT')
axes[0, 1].set_xlabel('Season')
axes[0, 1].set_ylabel('Temperature (°C)')
axes[0, 1].set_ylim(0, 45)
axes[0, 1].legend(loc='upper left')
axes[0, 1].grid(True)

# --- SSS & Precipitation Prior and Posterior (Shared Plot, Twin Axes) ---

# Get y-limits for SSS and precipitation (prior and posterior) to keep them consistent
sss_min = min(np.min(mu_prior_SSS_d18Ow_seasonal_T - stats.t.ppf(1 - 0.025, n_models_seasonal) * std_prior_SSS_d18Ow_seasonal_T / np.sqrt(n_models_seasonal)),
              np.min(mu_post_SSS_d18Ow_seasonal_aggregated_T - stats.t.ppf(1 - 0.025, n_update_seasonal_aggregated_D47 + n_models_seasonal) * std_post_SSS_d18Ow_seasonal_aggregated_T / np.sqrt(n_update_seasonal_aggregated_D47 + n_models_seasonal)))
sss_max = max(np.max(mu_prior_SSS_d18Ow_seasonal_T + stats.t.ppf(1 - 0.025, n_models_seasonal) * std_prior_SSS_d18Ow_seasonal_T / np.sqrt(n_models_seasonal)),
              np.max(mu_post_SSS_d18Ow_seasonal_aggregated_T + stats.t.ppf(1 - 0.025, n_update_seasonal_aggregated_D47 + n_models_seasonal) * std_post_SSS_d18Ow_seasonal_aggregated_T / np.sqrt(n_update_seasonal_aggregated_D47 + n_models_seasonal)))
precip_min = min(np.min(mu_prior_precip_seasonal_original - stats.t.ppf(1 - 0.025, n_models_seasonal) * std_prior_precip_seasonal / np.sqrt(n_models_seasonal)),
                 np.min(mu_post_precip_seasonal_aggregated - stats.t.ppf(1 - 0.025, n_update_seasonal_aggregated_D47 + n_models_seasonal) * np.sqrt(np.diag(cov_post_precip_seasonal_aggregated)) / np.sqrt(n_update_seasonal_aggregated_D47 + n_models_seasonal)))
precip_max = max(np.max(mu_prior_precip_seasonal_original + stats.t.ppf(1 - 0.025, n_models_seasonal) * std_prior_precip_seasonal / np.sqrt(n_models_seasonal)),
                 np.max(mu_post_precip_seasonal_aggregated + stats.t.ppf(1 - 0.025, n_update_seasonal_aggregated_D47 + n_models_seasonal) * np.sqrt(np.diag(cov_post_precip_seasonal_aggregated)) / np.sqrt(n_update_seasonal_aggregated_D47 + n_models_seasonal)))

# --- Prior SSS & Precipitation ---
ax1 = axes[1, 0]
ax2 = ax1.twinx()
lns1 = ax1.plot(seasons_scale, mu_prior_SSS_d18Ow_seasonal_T, label='Prior SSS Mean', marker='o', color="green")
fill1 = ax1.fill_between(
    seasons_scale,
    mu_prior_SSS_d18Ow_seasonal_T - stats.t.ppf(1 - 0.025, n_models_seasonal) * std_prior_SSS_d18Ow_seasonal_T / np.sqrt(n_models_seasonal),
    mu_prior_SSS_d18Ow_seasonal_T + stats.t.ppf(1 - 0.025, n_models_seasonal) * std_prior_SSS_d18Ow_seasonal_T / np.sqrt(n_models_seasonal),
    color="green", alpha=0.2, label='95% CI (SSS)'
)
lns2 = ax2.plot(seasons_scale, mu_prior_precip_seasonal_original, label='Prior Precip Mean', marker='o', color="purple")
fill2 = ax2.fill_between(
    seasons_scale,
    mu_prior_precip_seasonal_original - stats.t.ppf(1 - 0.025, n_models_seasonal) * std_prior_precip_seasonal / np.sqrt(n_models_seasonal),
    mu_prior_precip_seasonal_original + stats.t.ppf(1 - 0.025, n_models_seasonal) * std_prior_precip_seasonal / np.sqrt(n_models_seasonal),
    color="purple", alpha=0.2, label='95% CI (Precip)'
)
ax1.set_xticks(seasons_scale)
ax1.set_xticklabels(seasons)
ax1.set_title('Prior Mean and 95% CI for Seasonal SSS & Precipitation')
ax1.set_xlabel('Season')
ax1.set_ylabel('Salinity (PSU)', color="green")
ax2.set_ylabel('Precipitation (mm/day)', color="purple")
ax1.set_ylim(sss_min, sss_max)
ax2.set_ylim(precip_min, precip_max)
lns = lns1 + [fill1] + lns2 + [fill2]
labels = [l.get_label() for l in lns1] + [fill1.get_label()] + [l.get_label() for l in lns2] + [fill2.get_label()]
ax1.legend(lns, labels, loc='upper left')
ax1.grid(True)

# --- Posterior SSS & Precipitation ---
ax1 = axes[1, 1]
ax2 = ax1.twinx()
lns1 = ax1.plot(seasons_scale, mu_post_SSS_d18Ow_seasonal_aggregated_T, label='Posterior SSS Mean', marker='o', color="green")
fill1 = ax1.fill_between(
    seasons_scale,
    mu_post_SSS_d18Ow_seasonal_aggregated_T - stats.t.ppf(1 - 0.025, n_update_seasonal_aggregated_D47 + n_models_seasonal) * std_post_SSS_d18Ow_seasonal_aggregated_T / np.sqrt(n_update_seasonal_aggregated_D47 + n_models_seasonal),
    mu_post_SSS_d18Ow_seasonal_aggregated_T + stats.t.ppf(1 - 0.025, n_update_seasonal_aggregated_D47 + n_models_seasonal) * std_post_SSS_d18Ow_seasonal_aggregated_T / np.sqrt(n_update_seasonal_aggregated_D47 + n_models_seasonal),
    color="green", alpha=0.2, label='95% CI (SSS)'
)
lns2 = ax2.plot(seasons_scale, mu_post_precip_seasonal_aggregated, label='Posterior Precip Mean', marker='o', color="purple")
fill2 = ax2.fill_between(
    seasons_scale,
    mu_post_precip_seasonal_aggregated - stats.t.ppf(1 - 0.025, n_update_seasonal_aggregated_D47 + n_models_seasonal) * np.sqrt(np.diag(cov_post_precip_seasonal_aggregated)) / np.sqrt(n_update_seasonal_aggregated_D47 + n_models_seasonal),
    mu_post_precip_seasonal_aggregated + stats.t.ppf(1 - 0.025, n_update_seasonal_aggregated_D47 + n_models_seasonal) * np.sqrt(np.diag(cov_post_precip_seasonal_aggregated)) / np.sqrt(n_update_seasonal_aggregated_D47 + n_models_seasonal),
    color="purple", alpha=0.2, label='95% CI (Precip)'
)
ax1.set_xticks(seasons_scale)
ax1.set_xticklabels(seasons)
ax1.set_title('Posterior Mean and 95% CI for Seasonal SSS & Precipitation')
ax1.set_xlabel('Season')
ax1.set_ylabel('Salinity (PSU)', color="green")
ax2.set_ylabel('Precipitation (mm/day)', color="purple")
ax1.set_ylim(sss_min, sss_max)
ax2.set_ylim(precip_min, precip_max)
lns = lns1 + [fill1] + lns2 + [fill2]
labels = [l.get_label() for l in lns1] + [fill1.get_label()] + [l.get_label() for l in lns2] + [fill2.get_label()]
ax1.legend(lns, labels, loc='upper left')
ax1.grid(True)

# Add "n = x" labels below each x-tick for each subplot to show the number of model values or updates
# Prior SST & SAT
for i, season in enumerate(seasons):
    axes[0, 0].text(
        seasons_scale[i], axes[0, 0].get_ylim()[0] + 1.5,  # adjust location as needed
        f"n = {n_models_seasonal[i] if hasattr(n_models_seasonal, '__getitem__') else n_models_seasonal}",
        ha='center', va='top', fontsize=10
    )

# Posterior SST & SAT
for i, season in enumerate(seasons):
    axes[0, 1].text(
        seasons_scale[i], axes[0, 1].get_ylim()[0] + 1.5,
        f"n = {int(n_update_seasonal_aggregated_D47[i]) if hasattr(n_update_seasonal_aggregated_D47, '__getitem__') else n_update_seasonal_aggregated_D47}",
        ha='center', va='top', fontsize=10
    )

# Prior SSS & Precipitation
for i, season in enumerate(seasons):
    axes[1, 0].text(
        seasons_scale[i], axes[1, 0].get_ylim()[0] + 0.2,
        f"n = {n_models_seasonal[i] if hasattr(n_models_seasonal, '__getitem__') else n_models_seasonal}",
        ha='center', va='top', fontsize=10
    )

# Posterior SSS & Precipitation
for i, season in enumerate(seasons):
    axes[1, 1].text(
        seasons_scale[i], axes[1, 1].get_ylim()[0] + 0.2,
        f"n = {int(n_update_seasonal_aggregated_D47[i]) if hasattr(n_update_seasonal_aggregated_D47, '__getitem__') else n_update_seasonal_aggregated_D47}",
        ha='center', va='top', fontsize=10
    )

plt.tight_layout()
plt.show()

# Extract the number of datapoints for each season from Lutetian_data_dict
n_update_seasonal = np.array([len([d for d in Lutetian_data_dict if d['season_score'] == season]) for season in seasons_scale - 1])

# Create a figure with two rows and two columns of subplots
fig, axes = plt.subplots(2, 2, figsize=(18, 12))

# --- SST & SAT Prior and Posterior ---
# Prior SST & SAT
axes[0, 0].plot(seasons_scale, mu_prior_SST_seasonal, label='Prior SST Mean', marker='o')
axes[0, 0].plot(seasons_scale, mu_prior_SAT_seasonal, label='Prior SAT Mean', marker='o')
axes[0, 0].fill_between(
    seasons_scale,
    mu_prior_SST_seasonal - 2 * std_prior_SST_seasonal,
    mu_prior_SST_seasonal + 2 * std_prior_SST_seasonal,
    alpha=0.2, label='+/- 2 standard deviations (SST)'
)
axes[0, 0].fill_between(
    seasons_scale,
    mu_prior_SAT_seasonal - 2 * std_prior_SAT_seasonal,
    mu_prior_SAT_seasonal + 2 * std_prior_SAT_seasonal,
    alpha=0.2, label='+/- 2 standard deviations (SAT)'
)
axes[0, 0].set_xticks(seasons_scale)
axes[0, 0].set_xticklabels(seasons)
axes[0, 0].set_title('Prior Mean +/- 2 standard deviations for Seasonal SST & SAT')
axes[0, 0].set_xlabel('Season')
axes[0, 0].set_ylabel('Temperature (°C)')
axes[0, 0].set_ylim(0, 45)
axes[0, 0].legend(loc='upper left')
axes[0, 0].grid(True)

# Posterior SST & SAT
axes[0, 1].plot(seasons_scale, mu_post_SST_D47_seasonal_aggregated_T, label='Posterior SST Mean', marker='o')
axes[0, 1].plot(seasons_scale, mu_post_SAT_D47_seasonal_aggregated_T, label='Posterior SAT Mean', marker='o')
axes[0, 1].fill_between(
    seasons_scale,
    mu_post_SST_D47_seasonal_aggregated_T - 2 * std_post_SST_D47_seasonal_aggregated_T,
    mu_post_SST_D47_seasonal_aggregated_T + 2 * std_post_SST_D47_seasonal_aggregated_T,
    alpha=0.2, label='+/- 2 standard deviations (SST)'
)
axes[0, 1].fill_between(
    seasons_scale,
    mu_post_SAT_D47_seasonal_aggregated_T - 2 * std_post_SAT_D47_seasonal_aggregated_T,
    mu_post_SAT_D47_seasonal_aggregated_T + 2 * std_post_SAT_D47_seasonal_aggregated_T,
    alpha=0.2, label='+/- 2 standard deviations (SAT)'
)
axes[0, 1].set_xticks(seasons_scale)
axes[0, 1].set_xticklabels(seasons)
axes[0, 1].set_title('Posterior Mean +/- 2 standard deviations for Seasonal SST & SAT')
axes[0, 1].set_xlabel('Season')
axes[0, 1].set_ylabel('Temperature (°C)')
axes[0, 1].set_ylim(0, 45)
axes[0, 1].legend(loc='upper left')
axes[0, 1].grid(True)

# --- SSS & Precipitation Prior and Posterior (Shared Plot, Twin Axes) ---

# Get y-limits for SSS and precipitation (prior and posterior) to keep them consistent
sss_min = min(np.min(mu_prior_SSS_d18Ow_seasonal_T - 2 * std_prior_SSS_d18Ow_seasonal_T),
              np.min(mu_post_SSS_d18Ow_seasonal_aggregated_T - 2 * std_post_SSS_d18Ow_seasonal_aggregated_T))
sss_max = max(np.max(mu_prior_SSS_d18Ow_seasonal_T + 2 * std_prior_SSS_d18Ow_seasonal_T),
              np.max(mu_post_SSS_d18Ow_seasonal_aggregated_T + 2 * std_post_SSS_d18Ow_seasonal_aggregated_T))
precip_min = min(np.min(mu_prior_precip_seasonal_original - 2 * std_prior_precip_seasonal),
                 np.min(mu_post_precip_seasonal_aggregated - 2 * np.sqrt(np.diag(cov_post_precip_seasonal_aggregated))))
precip_max = max(np.max(mu_prior_precip_seasonal_original + 2 * std_prior_precip_seasonal),
                 np.max(mu_post_precip_seasonal_aggregated + 2 * np.sqrt(np.diag(cov_post_precip_seasonal_aggregated))))

# --- Prior SSS & Precipitation ---
ax1 = axes[1, 0]
ax2 = ax1.twinx()
lns1 = ax1.plot(seasons_scale, mu_prior_SSS_d18Ow_seasonal_T, label='Prior SSS Mean', marker='o', color="green")
fill1 = ax1.fill_between(
    seasons_scale,
    mu_prior_SSS_d18Ow_seasonal_T - 2 * std_prior_SSS_d18Ow_seasonal_T,
    mu_prior_SSS_d18Ow_seasonal_T + 2 * std_prior_SSS_d18Ow_seasonal_T,
    color="green", alpha=0.2, label='2 standard deviations (SSS)'
)
lns2 = ax2.plot(seasons_scale, mu_prior_precip_seasonal_original, label='Prior Precip Mean', marker='o', color="purple")
fill2 = ax2.fill_between(
    seasons_scale,
    mu_prior_precip_seasonal_original - 2 * std_prior_precip_seasonal,
    mu_prior_precip_seasonal_original + 2 * std_prior_precip_seasonal,
    color="purple", alpha=0.2, label='2 standard deviations (Precip)'
)
ax1.set_xticks(seasons_scale)
ax1.set_xticklabels(seasons)
ax1.set_title('Prior Mean +/- 2 standard deviations for Seasonal SSS & Precipitation')
ax1.set_xlabel('Season')
ax1.set_ylabel('Salinity (PSU)', color="green")
ax2.set_ylabel('Precipitation (mm/day)', color="purple")
ax1.set_ylim(sss_min, sss_max)
ax2.set_ylim(precip_min, precip_max)
lns = lns1 + [fill1] + lns2 + [fill2]
labels = [l.get_label() for l in lns1] + [fill1.get_label()] + [l.get_label() for l in lns2] + [fill2.get_label()]
ax1.legend(lns, labels, loc='upper left')
ax1.grid(True)

# --- Posterior SSS & Precipitation ---
ax1 = axes[1, 1]
ax2 = ax1.twinx()
lns1 = ax1.plot(seasons_scale, mu_post_SSS_d18Ow_seasonal_aggregated_T, label='Posterior SSS Mean', marker='o', color="green")
fill1 = ax1.fill_between(
    seasons_scale,
    mu_post_SSS_d18Ow_seasonal_aggregated_T - 2 * std_post_SSS_d18Ow_seasonal_aggregated_T,
    mu_post_SSS_d18Ow_seasonal_aggregated_T + 2 * std_post_SSS_d18Ow_seasonal_aggregated_T,
    color="green", alpha=0.2, label='2 standard deviations (SSS)'
)
lns2 = ax2.plot(seasons_scale, mu_post_precip_seasonal_aggregated, label='Posterior Precip Mean', marker='o', color="purple")
fill2 = ax2.fill_between(
    seasons_scale,
    mu_post_precip_seasonal_aggregated - 2 * np.sqrt(np.diag(cov_post_precip_seasonal_aggregated)),
    mu_post_precip_seasonal_aggregated + 2 * np.sqrt(np.diag(cov_post_precip_seasonal_aggregated)),
    color="purple", alpha=0.2, label='2 standard deviations (Precip)'
)
ax1.set_xticks(seasons_scale)
ax1.set_xticklabels(seasons)
ax1.set_title('Posterior Mean +/- 2 standard deviations for Seasonal SSS & Precipitation')
ax1.set_xlabel('Season')
ax1.set_ylabel('Salinity (PSU)', color="green")
ax2.set_ylabel('Precipitation (mm/day)', color="purple")
ax1.set_ylim(sss_min, sss_max)
ax2.set_ylim(precip_min, precip_max)
lns = lns1 + [fill1] + lns2 + [fill2]
labels = [l.get_label() for l in lns1] + [fill1.get_label()] + [l.get_label() for l in lns2] + [fill2.get_label()]
ax1.legend(lns, labels, loc='upper left')
ax1.grid(True)

# Add "n = x" labels below each x-tick for each subplot to show the number of model values or updates
# Prior SST & SAT
for i, season in enumerate(seasons):
    axes[0, 0].text(
        seasons_scale[i], axes[0, 0].get_ylim()[0] + 1.5,  # adjust location as needed
        f"n = {n_models_seasonal[i] if hasattr(n_models_seasonal, '__getitem__') else n_models_seasonal}",
        ha='center', va='top', fontsize=10
    )

# Posterior SST & SAT
for i, season in enumerate(seasons):
    axes[0, 1].text(
        seasons_scale[i], axes[0, 1].get_ylim()[0] + 1.5,
        f"n = {int(n_update_seasonal[i]) if hasattr(n_update_seasonal, '__getitem__') else n_update_seasonal}",
        ha='center', va='top', fontsize=10
    )

# Prior SSS & Precipitation
for i, season in enumerate(seasons):
    axes[1, 0].text(
        seasons_scale[i], axes[1, 0].get_ylim()[0] + 0.5,
        f"n = {n_models_seasonal[i] if hasattr(n_models_seasonal, '__getitem__') else n_models_seasonal}",
        ha='center', va='top', fontsize=10
    )

# Posterior SSS & Precipitation
for i, season in enumerate(seasons):
    axes[1, 1].text(
        seasons_scale[i], axes[1, 1].get_ylim()[0] + 0.5,
        f"n = {int(n_update_seasonal[i]) if hasattr(n_update_seasonal, '__getitem__') else n_update_seasonal}",
        ha='center', va='top', fontsize=10
    )

plt.tight_layout()
plt.show()

# Convert prior D47 to temp
mu_prior_SST_D47_monthly_T = D47c.OGLS23.T47(D47 = mu_prior_SST_D47_monthly_original, sD47 = cov_prior_SST_D47_monthly_original, return_covar = True)[0]
cov_prior_SST_D47_monthly_T = D47c.OGLS23.T47(D47 = mu_prior_SST_D47_monthly_original, sD47 = cov_prior_SST_D47_monthly_original, return_covar = True)[1]

# Convert prior d18Oc and temperature to d18Ow
mu_prior_d18Ow_monthly_T = mu_prior_d18Oc_monthly_original - (20.6 - mu_prior_SST_D47_monthly_T) / 4.34 + 0.27
# Calculate d18Oc-SST cross-covariance matrix
cross_cov_prior_d18Oc_SST_monthly = np.cov(Lutetian_models[SST_D47_columns_monthly + d18Oc_columns_monthly].dropna(), rowvar=False)[len(SST_D47_columns_monthly):, :len(SST_D47_columns_monthly)]
# Propagate covariance
cov_prior_d18Ow_monthly_T = propagate_cov_d18Ow_linear(cov_prior_d18Oc_monthly_original, cov_prior_SST_D47_monthly_T, cross_cov_prior_d18Oc_SST_monthly, 4.34)

# Convert prior d18Ow to SSS
mu_prior_SSS_d18Ow_monthly_T = (mu_prior_d18Ow_monthly_T + 9.300) / 0.274
cov_prior_SSS_d18Ow_monthly_T = cov_prior_d18Ow_monthly_T / (0.274 ** 2)

# Convert likelihood D47 to temp
# FIXME: likelihood vectors contain NAs, need to handle that properly
mu_likelihood_monthly_aggregated_T = D47c.OGLS23.T47(D47 = mu_likelihood_monthly_aggregated[var_start_D47_monthly:var_end_D47_monthly], sD47 = std_likelihood_monthly_aggregated[var_start_D47_monthly:var_end_D47_monthly], return_covar = True)[0]
cov_likelihood_monthly_aggregated_T = D47c.OGLS23.T47(D47 = mu_likelihood_monthly_aggregated[var_start_D47_monthly:var_end_D47_monthly], sD47 = std_likelihood_monthly_aggregated[var_start_D47_monthly:var_end_D47_monthly], return_covar = True)[1]

# Convert likelihood d18Oc and temperature to d18Ow (curently not implemented in data tracking function)
mu_likelihood_d18Ow_monthly_aggregated_T = mu_likelihood_monthly_aggregated[var_start_d18Oc_monthly:var_end_d18Oc_monthly] - (20.6 - mu_likelihood_monthly_aggregated_T) / 4.34 + 0.27
std_likelihood_d18Ow_monthly_aggregated_T = np.sqrt(
    np.diag(
        propagate_cov_d18Ow_linear(
            np.diag(std_likelihood_monthly_aggregated[var_start_d18Oc_monthly:var_end_d18Oc_monthly] ** 2),
            cov_likelihood_monthly_aggregated_T,
            cross_cov_prior_d18Oc_SST_monthly,
            4.34
        )
    )
)

# Convert likelihood d18Ow to SSS (currently not implemented in data tracking function)
mu_likelihood_SSS_d18Ow_monthly_aggregated_T = (mu_likelihood_d18Ow_monthly_aggregated_T + 9.300) / 0.274
std_likelihood_SSS_d18Ow_monthly_aggregated_T = std_likelihood_d18Ow_monthly_aggregated_T / (0.274 ** 2)

# Convert posterior D47 to temp
mu_post_SST_D47_monthly_aggregated_T = D47c.OGLS23.T47(D47 = mu_post_SST_D47_monthly_aggregated, sD47 = cov_post_SST_D47_monthly_aggregated, return_covar = True)[0]
cov_post_SST_D47_monthly_aggregated_T = D47c.OGLS23.T47(D47 = mu_post_SST_D47_monthly_aggregated, sD47 = cov_post_SST_D47_monthly_aggregated, return_covar = True)[1]

# Convert posterior d18Oc and temperature to d18Ow
mu_post_d18Ow_monthly_aggregated_T = mu_post_d18Oc_monthly_aggregated - (20.6 - mu_post_SST_D47_monthly_aggregated_T) / 4.34 + 0.27
cov_post_d18Ow_monthly_aggregated_T = propagate_cov_d18Ow_linear(
    cov_post_d18Oc_monthly_aggregated,
    cov_post_SST_D47_monthly_aggregated_T,
    cross_cov_prior_d18Oc_SST_monthly,
    4.34
)

# Convert posterior d18Ow to SSS
mu_post_SSS_d18Ow_monthly_aggregated_T = (mu_post_d18Ow_monthly_aggregated_T + 9.300) / 0.274
cov_post_SSS_d18Ow_monthly_aggregated_T = cov_post_d18Ow_monthly_aggregated_T / (0.274 ** 2)

# Plot in temperature domain
std_prior_SST_D47_monthly_T = np.sqrt(np.diag(cov_prior_SST_D47_monthly_T))
std_prior_SSS_d18Ow_monthly_T = np.sqrt(np.diag(cov_prior_SSS_d18Ow_monthly_T))
std_likelihood_monthly_aggregated_T = np.sqrt(np.diag(cov_likelihood_monthly_aggregated_T))
std_post_SST_D47_monthly_aggregated_T = np.sqrt(np.diag(cov_post_SST_D47_monthly_aggregated_T))
std_post_SSS_d18Ow_monthly_aggregated_T = np.sqrt(np.diag(cov_post_SSS_d18Ow_monthly_aggregated_T))

# Initiate plot
fig, axs = plt.subplots(2, 1, figsize=(10, 12), sharex=True)

# FIRST PANEL: SST Results
# PRIOR
axs[0].plot(months_scale, mu_prior_SST_D47_monthly_T, label='Prior Mean (PlioMIP models)', color='b', marker='o')
axs[0].fill_between(months_scale,
    mu_prior_SST_D47_monthly_T - stats.t.ppf(1 - 0.025, n_models_monthly) * std_prior_SST_D47_monthly_T / np.sqrt(n_models_monthly),
    mu_prior_SST_D47_monthly_T + stats.t.ppf(1 - 0.025, n_models_monthly) * std_prior_SST_D47_monthly_T / np.sqrt(n_models_monthly),
    color='b', alpha=0.2, label='95% Confidence Interval')

# LIKELIHOOD
axs[0].plot(months_scale, mu_likelihood_monthly_aggregated_T, label='Likelihood Mean (clumped data)', color='y', marker='o')
axs[0].fill_between(months_scale,
    mu_likelihood_monthly_aggregated_T - stats.t.ppf(1 - 0.025, n_update_monthly_aggregated_D47) * std_likelihood_monthly_aggregated_T / np.sqrt(n_update_monthly_aggregated_D47),
    mu_likelihood_monthly_aggregated_T + stats.t.ppf(1 - 0.025, n_update_monthly_aggregated_D47) * std_likelihood_monthly_aggregated_T / np.sqrt(n_update_monthly_aggregated_D47),
    color='y', alpha=0.2, label='95% Confidence Interval')

# POSTERIOR
axs[0].plot(months_scale, mu_post_SST_D47_monthly_aggregated_T, label='Posterior Mean (PlioMIP models + clumped data)', color='r', marker='o')
axs[0].fill_between(months_scale,
    mu_post_SST_D47_monthly_aggregated_T - stats.t.ppf(1 - 0.025, (n_update_monthly_aggregated_D47 + n_models_monthly)) * std_post_SST_D47_monthly_aggregated_T / np.sqrt(n_update_monthly_aggregated_D47 + n_models_monthly),
    mu_post_SST_D47_monthly_aggregated_T + stats.t.ppf(1 - 0.025, (n_update_monthly_aggregated_D47 + n_models_monthly)) * std_post_SST_D47_monthly_aggregated_T / np.sqrt(n_update_monthly_aggregated_D47 + n_models_monthly),
    color='r', alpha=0.2, label='95% Confidence Interval (Posterior)')

# Layout for SST panel
axs[0].set_title('monthly Sea Surface Temperatures (SST)')
axs[0].set_ylabel('Temperature (°C)')
axs[0].legend(loc='upper left')
axs[0].grid(True)

# SECOND PANEL: SSS Results
# PRIOR
axs[1].plot(months_scale, mu_prior_SSS_d18Ow_monthly_T, label='Prior Mean (PlioMIP models)', color='b', marker='o')
axs[1].fill_between(months_scale,
    mu_prior_SSS_d18Ow_monthly_T - stats.t.ppf(1 - 0.025, n_models_monthly) * std_prior_SSS_d18Ow_monthly_T / np.sqrt(n_models_monthly),
    mu_prior_SSS_d18Ow_monthly_T + stats.t.ppf(1 - 0.025, n_models_monthly) * std_prior_SSS_d18Ow_monthly_T / np.sqrt(n_models_monthly),
    color='b', alpha=0.2, label='95% Confidence Interval')

# LIKELIHOOD
axs[1].plot(months_scale, mu_likelihood_SSS_d18Ow_monthly_aggregated_T, label='Likelihood Mean (clumped data)', color='y', marker='o')
axs[1].fill_between(months_scale,
    mu_likelihood_SSS_d18Ow_monthly_aggregated_T - stats.t.ppf(1 - 0.025, n_update_monthly_aggregated_d18Oc) * std_likelihood_SSS_d18Ow_monthly_aggregated_T / np.sqrt(n_update_monthly_aggregated_d18Oc),
    mu_likelihood_SSS_d18Ow_monthly_aggregated_T + stats.t.ppf(1 - 0.025, n_update_monthly_aggregated_d18Oc) * std_likelihood_SSS_d18Ow_monthly_aggregated_T / np.sqrt(n_update_monthly_aggregated_d18Oc),
    color='y', alpha=0.2, label='95% Confidence Interval')

# POSTERIOR
axs[1].plot(months_scale, mu_post_SSS_d18Ow_monthly_aggregated_T, label='Posterior Mean (PlioMIP models + clumped data)', color='r', marker='o')
axs[1].fill_between(months_scale,
    mu_post_SSS_d18Ow_monthly_aggregated_T - stats.t.ppf(1 - 0.025, (n_update_monthly_aggregated_d18Oc + n_models_monthly)) * std_post_SSS_d18Ow_monthly_aggregated_T / np.sqrt(n_update_monthly_aggregated_d18Oc + n_models_monthly),
    mu_post_SSS_d18Ow_monthly_aggregated_T + stats.t.ppf(1 - 0.025, (n_update_monthly_aggregated_d18Oc + n_models_monthly)) * std_post_SSS_d18Ow_monthly_aggregated_T / np.sqrt(n_update_monthly_aggregated_d18Oc + n_models_monthly),
    color='r', alpha=0.2, label='95% Confidence Interval (Posterior)')

# Layout for SSS panel
axs[1].set_title('monthly Sea Surface Salinity (SSS)')
axs[1].set_xlabel('Month')
axs[1].set_ylabel('Salinity (PSU)')
axs[1].legend(loc='upper left')
axs[1].grid(True)

# Shared x-axis labels
plt.xticks(months_scale, month_names, rotation=45, ha="right")
plt.tight_layout()
plt.show()

# Convert posterior D47 to temp
mu_post_SAT_D47_monthly_aggregated_T = D47c.OGLS23.T47(D47 = mu_post_SAT_D47_monthly_aggregated, sD47 = cov_post_SAT_D47_monthly_aggregated, return_covar = True)[0]
cov_post_SAT_D47_monthly_aggregated_T = D47c.OGLS23.T47(D47 = mu_post_SAT_D47_monthly_aggregated, sD47 = cov_post_SAT_D47_monthly_aggregated, return_covar = True)[1]

# Convert posterior SAT-D47 back to temperature
std_post_SAT_D47_monthly_aggregated_T = np.nan_to_num(np.sqrt(np.diag(cov_post_SAT_D47_monthly_aggregated_T)))

fig, axs = plt.subplots(2, 1, figsize=(10, 12), sharex=True)

# --- SAT prior and posterior ---
# PRIOR
axs[0].plot(months_scale, mu_prior_SAT_monthly_original, label='Prior Mean (PlioMIP models)', color='b', marker='o')
axs[0].fill_between(
    months_scale,
    mu_prior_SAT_monthly_original - stats.t.ppf(1 - 0.025, n_models_monthly) * std_prior_SAT_monthly / np.sqrt(n_models_monthly),
    mu_prior_SAT_monthly_original + stats.t.ppf(1 - 0.025, n_models_monthly) * std_prior_SAT_monthly / np.sqrt(n_models_monthly),
    color='b', alpha=0.2, label='95% Confidence Interval'
)

# POSTERIOR
axs[0].plot(months_scale, mu_post_SAT_D47_monthly_aggregated_T, label='Posterior Mean (PlioMIP models + clumped data)', color='r', marker='o')
axs[0].fill_between(
    months_scale,
    mu_post_SAT_D47_monthly_aggregated_T - stats.t.ppf(1 - 0.025, (n_update_monthly_aggregated_D47 + n_models_monthly)) * std_post_SAT_D47_monthly_aggregated_T / np.sqrt(n_update_monthly_aggregated_D47 + n_models_monthly),
    mu_post_SAT_D47_monthly_aggregated_T + stats.t.ppf(1 - 0.025, (n_update_monthly_aggregated_D47 + n_models_monthly)) * std_post_SAT_D47_monthly_aggregated_T / np.sqrt(n_update_monthly_aggregated_D47 + n_models_monthly),
    color='r', alpha=0.2, label='95% Confidence Interval (Posterior)'
)

axs[0].set_title('Posterior Mean and 95% Confidence Interval for monthly Surface Air Temperatures\n(Based on monthly averages per specimen)')
axs[0].set_ylabel('Temperature (°C)')
axs[0].set_ylim(0, 50)
axs[0].legend(loc='upper left')
axs[0].grid(True)

# --- Precipitation prior and posterior ---
# PRIOR
axs[1].plot(months_scale, mu_prior_precip_monthly_original, label='Prior Mean (PlioMIP models)', color='b', marker='o')
axs[1].fill_between(
    months_scale,
    mu_prior_precip_monthly_original - stats.t.ppf(1 - 0.025, n_models_monthly) * std_prior_precip_monthly / np.sqrt(n_models_monthly),
    mu_prior_precip_monthly_original + stats.t.ppf(1 - 0.025, n_models_monthly) * std_prior_precip_monthly / np.sqrt(n_models_monthly),
    color='b', alpha=0.2, label='95% Confidence Interval'
)

# POSTERIOR
axs[1].plot(months_scale, mu_post_precip_monthly_aggregated, label='Posterior Mean (PlioMIP models + clumped data)', color='r', marker='o')
axs[1].fill_between(
    months_scale,
    mu_post_precip_monthly_aggregated - stats.t.ppf(1 - 0.025, (n_update_monthly_aggregated_D47 + n_models_monthly)) * np.sqrt(np.diag(cov_post_precip_monthly_aggregated)) / np.sqrt(n_update_monthly_aggregated_D47 + n_models_monthly),
    mu_post_precip_monthly_aggregated + stats.t.ppf(1 - 0.025, (n_update_monthly_aggregated_D47 + n_models_monthly)) * np.sqrt(np.diag(cov_post_precip_monthly_aggregated)) / np.sqrt(n_update_monthly_aggregated_D47 + n_models_monthly),
    color='r', alpha=0.2, label='95% Confidence Interval (Posterior)'
)

axs[1].set_title('Posterior Mean and 95% Confidence Interval for monthly Precipitation')
axs[1].set_xlabel('Month')
axs[1].set_ylabel('Precipitation (mm/day)')
axs[1].legend(loc='upper left')
axs[1].grid(True)

plt.xticks(months_scale, month_names, rotation=45, ha="right")
plt.tight_layout()
plt.show()

# Create a figure with two rows and two columns of subplots
fig, axes = plt.subplots(2, 2, figsize=(18, 12))

# --- SST & SAT Prior and Posterior ---
# Prior SST & SAT
axes[0, 0].plot(months_scale, mu_prior_SST_monthly, label='Prior SST Mean', marker='o')
axes[0, 0].plot(months_scale, mu_prior_SAT_monthly, label='Prior SAT Mean', marker='o')
axes[0, 0].fill_between(
    months_scale,
    mu_prior_SST_monthly - stats.t.ppf(1 - 0.025, n_models_monthly) * std_prior_SST_monthly / np.sqrt(n_models_monthly),
    mu_prior_SST_monthly + stats.t.ppf(1 - 0.025, n_models_monthly) * std_prior_SST_monthly / np.sqrt(n_models_monthly),
    alpha=0.2, label='95% Confidence Interval (SST)'
)
axes[0, 0].fill_between(
    months_scale,
    mu_prior_SAT_monthly - stats.t.ppf(1 - 0.025, n_models_monthly) * std_prior_SAT_monthly / np.sqrt(n_models_monthly),
    mu_prior_SAT_monthly + stats.t.ppf(1 - 0.025, n_models_monthly) * std_prior_SAT_monthly / np.sqrt(n_models_monthly),
    alpha=0.2, label='95% Confidence Interval (SAT)'
)
axes[0, 0].set_xticks(months_scale)
axes[0, 0].set_xticklabels(month_names, rotation=45, ha="right")
axes[0, 0].set_title('Prior Mean and 95% CI for monthly SST & SAT')
axes[0, 0].set_xlabel('Month')
axes[0, 0].set_ylabel('Temperature (°C)')
axes[0, 0].set_ylim(0, 50)
axes[0, 0].legend(loc='upper left')
axes[0, 0].grid(True)

# Posterior SST & SAT
axes[0, 1].plot(months_scale, mu_post_SST_D47_monthly_aggregated_T, label='Posterior SST Mean', marker='o')
axes[0, 1].plot(months_scale, mu_post_SAT_D47_monthly_aggregated_T, label='Posterior SAT Mean', marker='o')
axes[0, 1].fill_between(
    months_scale,
    mu_post_SST_D47_monthly_aggregated_T - stats.t.ppf(1 - 0.025, n_update_monthly_aggregated_D47 + n_models_monthly) * std_post_SST_D47_monthly_aggregated_T / np.sqrt(n_update_monthly_aggregated_D47 + n_models_monthly),
    mu_post_SST_D47_monthly_aggregated_T + stats.t.ppf(1 - 0.025, n_update_monthly_aggregated_D47 + n_models_monthly) * std_post_SST_D47_monthly_aggregated_T / np.sqrt(n_update_monthly_aggregated_D47 + n_models_monthly),
    alpha=0.2, label='95% Confidence Interval (SST)'
)
axes[0, 1].fill_between(
    months_scale,
    mu_post_SAT_D47_monthly_aggregated_T - stats.t.ppf(1 - 0.025, n_update_monthly_aggregated_D47 + n_models_monthly) * std_post_SAT_D47_monthly_aggregated_T / np.sqrt(n_update_monthly_aggregated_D47 + n_models_monthly),
    mu_post_SAT_D47_monthly_aggregated_T + stats.t.ppf(1 - 0.025, n_update_monthly_aggregated_D47 + n_models_monthly) * std_post_SAT_D47_monthly_aggregated_T / np.sqrt(n_update_monthly_aggregated_D47 + n_models_monthly),
    alpha=0.2, label='95% Confidence Interval (SAT)'
)
axes[0, 1].set_xticks(months_scale)
axes[0, 1].set_xticklabels(month_names, rotation=45, ha="right")
axes[0, 1].set_title('Posterior Mean and 95% CI for monthly SST & SAT')
axes[0, 1].set_xlabel('Month')
axes[0, 1].set_ylabel('Temperature (°C)')
axes[0, 1].set_ylim(0, 50)
axes[0, 1].legend(loc='upper left')
axes[0, 1].grid(True)

# --- SSS & Precipitation Prior and Posterior (Shared Plot, Twin Axes) ---

# Get y-limits for SSS and precipitation (prior and posterior) to keep them consistent
sss_min = min(np.min(mu_prior_SSS_d18Ow_monthly_T - stats.t.ppf(1 - 0.025, n_models_monthly) * std_prior_SSS_d18Ow_monthly_T / np.sqrt(n_models_monthly)),
              np.min(mu_post_SSS_d18Ow_monthly_aggregated_T - stats.t.ppf(1 - 0.025, n_update_monthly_aggregated_D47 + n_models_monthly) * std_post_SSS_d18Ow_monthly_aggregated_T / np.sqrt(n_update_monthly_aggregated_D47 + n_models_monthly)))
sss_max = max(np.max(mu_prior_SSS_d18Ow_monthly_T + stats.t.ppf(1 - 0.025, n_models_monthly) * std_prior_SSS_d18Ow_monthly_T / np.sqrt(n_models_monthly)),
              np.max(mu_post_SSS_d18Ow_monthly_aggregated_T + stats.t.ppf(1 - 0.025, n_update_monthly_aggregated_D47 + n_models_monthly) * std_post_SSS_d18Ow_monthly_aggregated_T / np.sqrt(n_update_monthly_aggregated_D47 + n_models_monthly)))
precip_min = min(np.min(mu_prior_precip_monthly_original - stats.t.ppf(1 - 0.025, n_models_monthly) * std_prior_precip_monthly / np.sqrt(n_models_monthly)),
                 np.min(mu_post_precip_monthly_aggregated - stats.t.ppf(1 - 0.025, n_update_monthly_aggregated_D47 + n_models_monthly) * np.sqrt(np.diag(cov_post_precip_monthly_aggregated)) / np.sqrt(n_update_monthly_aggregated_D47 + n_models_monthly)))
precip_max = max(np.max(mu_prior_precip_monthly_original + stats.t.ppf(1 - 0.025, n_models_monthly) * std_prior_precip_monthly / np.sqrt(n_models_monthly)),
                 np.max(mu_post_precip_monthly_aggregated + stats.t.ppf(1 - 0.025, n_update_monthly_aggregated_D47 + n_models_monthly) * np.sqrt(np.diag(cov_post_precip_monthly_aggregated)) / np.sqrt(n_update_monthly_aggregated_D47 + n_models_monthly)))

# --- Prior SSS & Precipitation ---
ax1 = axes[1, 0]
ax2 = ax1.twinx()
lns1 = ax1.plot(months_scale, mu_prior_SSS_d18Ow_monthly_T, label='Prior SSS Mean', marker='o', color="green")
fill1 = ax1.fill_between(
    months_scale,
    mu_prior_SSS_d18Ow_monthly_T - stats.t.ppf(1 - 0.025, n_models_monthly) * std_prior_SSS_d18Ow_monthly_T / np.sqrt(n_models_monthly),
    mu_prior_SSS_d18Ow_monthly_T + stats.t.ppf(1 - 0.025, n_models_monthly) * std_prior_SSS_d18Ow_monthly_T / np.sqrt(n_models_monthly),
    color="green", alpha=0.2, label='95% CI (SSS)'
)
lns2 = ax2.plot(months_scale, mu_prior_precip_monthly_original, label='Prior Precip Mean', marker='o', color="purple")
fill2 = ax2.fill_between(
    months_scale,
    mu_prior_precip_monthly_original - stats.t.ppf(1 - 0.025, n_models_monthly) * std_prior_precip_monthly / np.sqrt(n_models_monthly),
    mu_prior_precip_monthly_original + stats.t.ppf(1 - 0.025, n_models_monthly) * std_prior_precip_monthly / np.sqrt(n_models_monthly),
    color="purple", alpha=0.2, label='95% CI (Precip)'
)
ax1.set_xticks(months_scale)
ax1.set_xticklabels(month_names, rotation=45, ha="right")
ax1.set_title('Prior Mean and 95% CI for monthly SSS & Precipitation')
ax1.set_xlabel('Month')
ax1.set_ylabel('Salinity (PSU)', color="green")
ax2.set_ylabel('Precipitation (mm/day)', color="purple")
ax1.set_ylim(sss_min, sss_max)
ax2.set_ylim(precip_min, precip_max)
lns = lns1 + [fill1] + lns2 + [fill2]
labels = [l.get_label() for l in lns1] + [fill1.get_label()] + [l.get_label() for l in lns2] + [fill2.get_label()]
ax1.legend(lns, labels, loc='upper left')
ax1.grid(True)

# --- Posterior SSS & Precipitation ---
ax1 = axes[1, 1]
ax2 = ax1.twinx()
lns1 = ax1.plot(months_scale, mu_post_SSS_d18Ow_monthly_aggregated_T, label='Posterior SSS Mean', marker='o', color="green")
fill1 = ax1.fill_between(
    months_scale,
    mu_post_SSS_d18Ow_monthly_aggregated_T - stats.t.ppf(1 - 0.025, n_update_monthly_aggregated_D47 + n_models_monthly) * std_post_SSS_d18Ow_monthly_aggregated_T / np.sqrt(n_update_monthly_aggregated_D47 + n_models_monthly),
    mu_post_SSS_d18Ow_monthly_aggregated_T + stats.t.ppf(1 - 0.025, n_update_monthly_aggregated_D47 + n_models_monthly) * std_post_SSS_d18Ow_monthly_aggregated_T / np.sqrt(n_update_monthly_aggregated_D47 + n_models_monthly),
    color="green", alpha=0.2, label='95% CI (SSS)'
)
lns2 = ax2.plot(months_scale, mu_post_precip_monthly_aggregated, label='Posterior Precip Mean', marker='o', color="purple")
fill2 = ax2.fill_between(
    months_scale,
    mu_post_precip_monthly_aggregated - stats.t.ppf(1 - 0.025, n_update_monthly_aggregated_D47 + n_models_monthly) * np.sqrt(np.diag(cov_post_precip_monthly_aggregated)) / np.sqrt(n_update_monthly_aggregated_D47 + n_models_monthly),
    mu_post_precip_monthly_aggregated + stats.t.ppf(1 - 0.025, n_update_monthly_aggregated_D47 + n_models_monthly) * np.sqrt(np.diag(cov_post_precip_monthly_aggregated)) / np.sqrt(n_update_monthly_aggregated_D47 + n_models_monthly),
    color="purple", alpha=0.2, label='95% CI (Precip)'
)
ax1.set_xticks(months_scale)
ax1.set_xticklabels(month_names, rotation=45, ha="right")
ax1.set_title('Posterior Mean and 95% CI for monthly SSS & Precipitation')
ax1.set_xlabel('Month')
ax1.set_ylabel('Salinity (PSU)', color="green")
ax2.set_ylabel('Precipitation (mm/day)', color="purple")
ax1.set_ylim(sss_min, sss_max)
ax2.set_ylim(precip_min, precip_max)
lns = lns1 + [fill1] + lns2 + [fill2]
labels = [l.get_label() for l in lns1] + [fill1.get_label()] + [l.get_label() for l in lns2] + [fill2.get_label()]
ax1.legend(lns, labels, loc='upper left')
ax1.grid(True)

# Add "n = x" labels below each x-tick for each subplot to show the number of model values or updates
# Prior SST & SAT
for i, month in enumerate(month_names):
    axes[0, 0].text(
        months_scale[i], axes[0, 0].get_ylim()[0] + 1.5,  # adjust location as needed
        f"n = {n_models_monthly[i] if hasattr(n_models_monthly, '__getitem__') else n_models_monthly}",
        ha='center', va='top', fontsize=10
    )

# Posterior SST & SAT
for i, month in enumerate(month_names):
    axes[0, 1].text(
        months_scale[i], axes[0, 1].get_ylim()[0] + 1.5,
        f"n = {int(n_update_monthly_aggregated_D47[i]) if hasattr(n_update_monthly_aggregated_D47, '__getitem__') else n_update_monthly_aggregated_D47}",
        ha='center', va='top', fontsize=10
    )

# Prior SSS & Precipitation
for i, month in enumerate(month_names):
    axes[1, 0].text(
        months_scale[i], axes[1, 0].get_ylim()[0] + 0.5,
        f"n = {n_models_monthly[i] if hasattr(n_models_monthly, '__getitem__') else n_models_monthly}",
        ha='center', va='top', fontsize=10
    )

# Posterior SSS & Precipitation
for i, month in enumerate(month_names):
    axes[1, 1].text(
        months_scale[i], axes[1, 1].get_ylim()[0] + 0.5,
        f"n = {int(n_update_monthly_aggregated_D47[i]) if hasattr(n_update_monthly_aggregated_D47, '__getitem__') else n_update_monthly_aggregated_D47}",
        ha='center', va='top', fontsize=10
    )
    
plt.tight_layout()
plt.show()

# Extract the number of datapoints for each month from Lutetian_data_dict
n_update_monthly = np.array([len([d for d in Lutetian_data_dict if d['month_score'] == month]) for month in months_scale - 1])

# Create a figure with two rows and two columns of subplots
fig, axes = plt.subplots(2, 2, figsize=(18, 12))

# --- SST & SAT Prior and Posterior ---
# Prior SST & SAT
axes[0, 0].plot(months_scale, mu_prior_SST_monthly, label='Prior SST Mean', marker='o')
axes[0, 0].plot(months_scale, mu_prior_SAT_monthly, label='Prior SAT Mean', marker='o')
axes[0, 0].fill_between(
    months_scale,
    mu_prior_SST_monthly - 2 * std_prior_SST_monthly,
    mu_prior_SST_monthly + 2 * std_prior_SST_monthly,
    alpha=0.2, label='+/- 2 standard deviations (SST)'
)
axes[0, 0].fill_between(
    months_scale,
    mu_prior_SAT_monthly - 2 * std_prior_SAT_monthly,
    mu_prior_SAT_monthly + 2 * std_prior_SAT_monthly,
    alpha=0.2, label='+/- 2 standard deviations (SAT)'
)
axes[0, 0].set_xticks(months_scale)
axes[0, 0].set_xticklabels(month_names, rotation=45, ha="right")
axes[0, 0].set_title('Prior Mean +/- 2 standard deviations for monthly SST & SAT')
axes[0, 0].set_xlabel('Month')
axes[0, 0].set_ylabel('Temperature (°C)')
axes[0, 0].set_ylim(0, 50)
axes[0, 0].legend(loc='upper left')
axes[0, 0].grid(True)

# Posterior SST & SAT
axes[0, 1].plot(months_scale, mu_post_SST_D47_monthly_aggregated_T, label='Posterior SST Mean', marker='o')
axes[0, 1].plot(months_scale, mu_post_SAT_D47_monthly_aggregated_T, label='Posterior SAT Mean', marker='o')
axes[0, 1].fill_between(
    months_scale,
    mu_post_SST_D47_monthly_aggregated_T - 2 * std_post_SST_D47_monthly_aggregated_T,
    mu_post_SST_D47_monthly_aggregated_T + 2 * std_post_SST_D47_monthly_aggregated_T,
    alpha=0.2, label='+/- 2 standard deviations (SST)'
)
axes[0, 1].fill_between(
    months_scale,
    mu_post_SAT_D47_monthly_aggregated_T - 2 * std_post_SAT_D47_monthly_aggregated_T,
    mu_post_SAT_D47_monthly_aggregated_T + 2 * std_post_SAT_D47_monthly_aggregated_T,
    alpha=0.2, label='+/- 2 standard deviations (SAT)'
)
axes[0, 1].set_xticks(months_scale)
axes[0, 1].set_xticklabels(month_names, rotation=45, ha="right")
axes[0, 1].set_title('Posterior Mean +/- 2 standard deviations for monthly SST & SAT')
axes[0, 1].set_xlabel('Month')
axes[0, 1].set_ylabel('Temperature (°C)')
axes[0, 1].set_ylim(0, 50)
axes[0, 1].legend(loc='upper left')
axes[0, 1].grid(True)

# --- SSS & Precipitation Prior and Posterior (Shared Plot, Twin Axes) ---

# Get y-limits for SSS and precipitation (prior and posterior) to keep them consistent
sss_min = min(np.min(mu_prior_SSS_d18Ow_monthly_T - 2 * std_prior_SSS_d18Ow_monthly_T),
              np.min(mu_post_SSS_d18Ow_monthly_aggregated_T - 2 * std_post_SSS_d18Ow_monthly_aggregated_T))
sss_max = max(np.max(mu_prior_SSS_d18Ow_monthly_T + 2 * std_prior_SSS_d18Ow_monthly_T),
              np.max(mu_post_SSS_d18Ow_monthly_aggregated_T + 2 * std_post_SSS_d18Ow_monthly_aggregated_T))
precip_min = min(np.min(mu_prior_precip_monthly_original - 2 * std_prior_precip_monthly),
                 np.min(mu_post_precip_monthly_aggregated - 2 * np.sqrt(np.diag(cov_post_precip_monthly_aggregated))))
precip_max = max(np.max(mu_prior_precip_monthly_original + 2 * std_prior_precip_monthly),
                 np.max(mu_post_precip_monthly_aggregated + 2 * np.sqrt(np.diag(cov_post_precip_monthly_aggregated))))

# --- Prior SSS & Precipitation ---
ax1 = axes[1, 0]
ax2 = ax1.twinx()
lns1 = ax1.plot(months_scale, mu_prior_SSS_d18Ow_monthly_T, label='Prior SSS Mean', marker='o', color="green")
fill1 = ax1.fill_between(
    months_scale,
    mu_prior_SSS_d18Ow_monthly_T - 2 * std_prior_SSS_d18Ow_monthly_T,
    mu_prior_SSS_d18Ow_monthly_T + 2 * std_prior_SSS_d18Ow_monthly_T,
    color="green", alpha=0.2, label='2 standard deviations (SSS)'
)
lns2 = ax2.plot(months_scale, mu_prior_precip_monthly_original, label='Prior Precip Mean', marker='o', color="purple")
fill2 = ax2.fill_between(
    months_scale,
    mu_prior_precip_monthly_original - 2 * std_prior_precip_monthly,
    mu_prior_precip_monthly_original + 2 * std_prior_precip_monthly,
    color="purple", alpha=0.2, label='2 standard deviations (Precip)'
)
ax1.set_xticks(months_scale)
ax1.set_xticklabels(month_names, rotation=45, ha="right")
ax1.set_title('Prior Mean +/- 2 standard deviations for monthly SSS & Precipitation')
ax1.set_xlabel('Month')
ax1.set_ylabel('Salinity (PSU)', color="green")
ax2.set_ylabel('Precipitation (mm/day)', color="purple")
ax1.set_ylim(sss_min, sss_max)
ax2.set_ylim(precip_min, precip_max)
lns = lns1 + [fill1] + lns2 + [fill2]
labels = [l.get_label() for l in lns1] + [fill1.get_label()] + [l.get_label() for l in lns2] + [fill2.get_label()]
ax1.legend(lns, labels, loc='upper left')
ax1.grid(True)

# --- Posterior SSS & Precipitation ---
ax1 = axes[1, 1]
ax2 = ax1.twinx()
lns1 = ax1.plot(months_scale, mu_post_SSS_d18Ow_monthly_aggregated_T, label='Posterior SSS Mean', marker='o', color="green")
fill1 = ax1.fill_between(
    months_scale,
    mu_post_SSS_d18Ow_monthly_aggregated_T - 2 * std_post_SSS_d18Ow_monthly_aggregated_T,
    mu_post_SSS_d18Ow_monthly_aggregated_T + 2 * std_post_SSS_d18Ow_monthly_aggregated_T,
    color="green", alpha=0.2, label='2 standard deviations (SSS)'
)
lns2 = ax2.plot(months_scale, mu_post_precip_monthly_aggregated, label='Posterior Precip Mean', marker='o', color="purple")
fill2 = ax2.fill_between(
    months_scale,
    mu_post_precip_monthly_aggregated - 2 * np.sqrt(np.diag(cov_post_precip_monthly_aggregated)),
    mu_post_precip_monthly_aggregated + 2 * np.sqrt(np.diag(cov_post_precip_monthly_aggregated)),
    color="purple", alpha=0.2, label='2 standard deviations (Precip)'
)
ax1.set_xticks(months_scale)
ax1.set_xticklabels(month_names, rotation=45, ha="right")
ax1.set_title('Posterior Mean +/- 2 standard deviations for monthly SSS & Precipitation')
ax1.set_xlabel('Month')
ax1.set_ylabel('Salinity (PSU)', color="green")
ax2.set_ylabel('Precipitation (mm/day)', color="purple")
ax1.set_ylim(sss_min, sss_max)
ax2.set_ylim(precip_min, precip_max)
lns = lns1 + [fill1] + lns2 + [fill2]
labels = [l.get_label() for l in lns1] + [fill1.get_label()] + [l.get_label() for l in lns2] + [fill2.get_label()]
ax1.legend(lns, labels, loc='upper left')
ax1.grid(True)

# Add "n = x" labels below each x-tick for each subplot to show the number of model values or updates
# Prior SST & SAT
for i, month in enumerate(month_names):
    axes[0, 0].text(
        months_scale[i], axes[0, 0].get_ylim()[0] + 1.5,  # adjust location as needed
        f"n = {n_models_monthly[i] if hasattr(n_models_monthly, '__getitem__') else n_models_monthly}",
        ha='center', va='top', fontsize=10
    )

# Posterior SST & SAT
for i, month in enumerate(month_names):
    axes[0, 1].text(
        months_scale[i], axes[0, 1].get_ylim()[0] + 1.5,
        f"n = {int(n_update_monthly[i]) if hasattr(n_update_monthly, '__getitem__') else n_update_monthly}",
        ha='center', va='top', fontsize=10
    )

# Prior SSS & Precipitation
for i, month in enumerate(month_names):
    axes[1, 0].text(
        months_scale[i], axes[1, 0].get_ylim()[0] + 0.5,
        f"n = {n_models_monthly[i] if hasattr(n_models_monthly, '__getitem__') else n_models_monthly}",
        ha='center', va='top', fontsize=10
    )

# Posterior SSS & Precipitation
for i, month in enumerate(month_names):
    axes[1, 1].text(
        months_scale[i], axes[1, 1].get_ylim()[0] + 0.5,
        f"n = {int(n_update_monthly[i]) if hasattr(n_update_monthly, '__getitem__') else n_update_monthly}",
        ha='center', va='top', fontsize=10
    )
    
plt.tight_layout()
plt.show()

	Month	Temperature
0	1	11.282648
1	1	21.823206
2	1	21.909296
3	1	22.503198
4	1	21.393762

	Month	SST
0	1	25.653111
1	1	25.308791
2	1	24.928478
3	1	18.123590
4	1	18.129648

	Cell	ja_SAT	fb_SAT	mr_SAT	ar_SAT	my_SAT	jn_SAT	jl_SAT	ag_SAT	sp_SAT	...	mr_SSS	ar_SSS	my_SSS	jn_SSS	jl_SSS	ag_SSS	sp_SSS	ot_SSS	nv_SSS	dc_SSS
0	1	11.282648	12.089380	13.829187	16.709039	22.411493	27.940820	30.968195	31.072290	26.694391	...	35.461410	35.476315	35.502321	35.524816	35.519439	35.469567	35.353144	35.262145	35.298730	35.386242
1	2	21.823206	22.185327	23.480707	25.735864	30.242242	35.107660	38.680688	39.036523	35.629755	...	35.474837	35.491815	35.518168	35.535836	35.529558	35.499538	35.409034	35.322384	35.326226	35.393769
2	3	21.909296	22.198022	23.222040	25.234583	29.434015	34.056543	38.089227	38.691492	35.554956	...	35.391840	35.407201	35.434341	35.446876	35.436236	35.413111	35.338840	35.276135	35.276722	35.330491
3	4	22.503198	22.530481	23.346246	25.152704	28.970605	33.480951	37.780298	38.492792	35.470605	...	27.740462	27.669306	27.565980	27.477787	27.376960	27.298817	27.239179	27.249114	27.380852	27.560293
4	5	21.393762	21.576868	22.996118	25.669275	30.826440	36.038263	40.872888	40.882013	36.701257	...	27.695703	27.609018	27.476092	27.362065	27.253544	27.175085	27.119611	27.145849	27.320097	27.518127

	Cell	ja_SAT	fb_SAT	mr_SAT	ar_SAT	my_SAT	jn_SAT	jl_SAT	ag_SAT	sp_SAT	...	mr_d18Oc	ar_d18Oc	my_d18Oc	jn_d18Oc	jl_d18Oc	ag_d18Oc	sp_d18Oc	ot_d18Oc	nv_d18Oc	dc_d18Oc
0	1	11.282648	12.089380	13.829187	16.709039	22.411493	27.940820	30.968195	31.072290	26.694391	...	-0.873077	-1.040673	-1.619817	-2.513306	-3.213928	-3.583698	-3.367792	-2.744058	-2.032903	-1.408096
1	2	21.823206	22.185327	23.480707	25.735864	30.242242	35.107660	38.680688	39.036523	35.629755	...	-0.799470	-0.929937	-1.466061	-2.347593	-3.014437	-3.384486	-3.208780	-2.633968	-1.938649	-1.313754
2	3	21.909296	22.198022	23.222040	25.234583	29.434015	34.056543	38.089227	38.691492	35.554956	...	-0.754007	-0.877557	-1.403951	-2.289134	-2.948919	-3.317429	-3.148663	-2.580306	-1.883936	-1.240408
3	4	22.503198	22.530481	23.346246	25.152704	28.970605	33.480951	37.780298	38.492792	35.470605	...	-1.136514	-1.384207	-2.130491	-3.116770	-3.959694	-4.334650	-4.195826	-3.656621	-2.769984	-1.943248
4	5	21.393762	21.576868	22.996118	25.669275	30.826440	36.038263	40.872888	40.882013	36.701257	...	-1.122320	-1.360099	-2.106644	-3.107112	-3.975039	-4.357027	-4.219610	-3.656945	-2.755215	-1.950705

INTRO¶

Notebook for applying Data Assimilation on seasonal data (Lutetian reconstructions)¶

Assemble data in the proxy domain using offline (block update) assimilation instead of point-by-point¶

Assemble data in SST (D47), SAT (D47), SSS (d18Oc) and precipitation domain¶

Load packages¶

PRIOR - MONTHLY¶

Load monthly SAT model data¶

Load monthly SST model data¶

Load monthly SSS model data¶

Load monthly precipitation model data¶

Combine SAT and precipitation data by modelname¶

Calculate the monthly prior for model SST, SAT, SSS and precipitation values¶

Plot the monthly priors for all model values¶

Convert SAT and SST model data to D47 domain using the regression by Daëron and Vermeesch (2023) and propagate uncertainty in the calibration¶

Estimate seawater oxygen isotope value from salinity based on modern North Sea d18Ow-salinity relationship by Harwood et al. (2007)¶

Calculate carbonate oxygen isotope value from SST and seawater oxygen isotope data using Grossman and Ku (1986) with the VPDB-VSMOW scale correction by Gonfiantini et al. (1995) and Dettman et al. (1999)¶

Calculate the monthly prior for model SST- and SAT-derived D47 values and SSS-derived seawater oxygen isotope values with propagated uncertainty¶

Plot the monthly prior for model SST- and SAT-derived D47 values, model SSS-derived carbonate d18O values and precipitation with propagated uncertainty¶

Calculate the monthly covariance matrix for D47 values of SST and SAT, d18Oc and precipitation¶

Plot normalized monthly covariance matrix between D47 values of SST and SAT, d18Oc and precipitation¶

Create combined monthly state vector¶

PRIOR - SEASONAL¶

Seasonal model data (convert the model data to seasonal means)¶

Calculate the seasonal prior for model SST- and SAT-derived D47 values and SSS-derived seawater oxygen isotope values with propagated uncertainty¶

Plot the seasonal prior for model SST, SAT, SSS and precipitation¶

Calculate the seasonal prior for model SST and SAT-derived D47 values with propagated uncertainty¶

Plot the seasonal prior for model SST- and SAT-derived D47 values, d18Oc values and precipitation with propagated uncertainty¶

Calculate the seasonal covariance matrix for D47 values derived from SST and SAT values, d18Oc and precipitation¶

Plot normalized seasonal covariance matrix between D47 values of SST and SAT, d18Oc and precipitation¶

Create combined seasonal state vector¶

OBSERVATIONS¶

Load clumped data for updating¶

Monthly data from Paris Basin Campanile giganteum paper (Van Horebeek et al. 2025)¶

Raw data at the sample level¶

Monthly and seasonal data calculated from ShellChron outcomes¶

Aggregate proxy data to seasonal and monthly bins¶

Prepare measurement and observation matrices¶

Define a wrapped normal distribution to allow uncertainty in the time domain to flow around the year¶

Observations aggregated by season¶

Measurement matrix for season- and monthly-averaged D47 and d18Oc values¶

Observation matrix for season- and monthly-averaged D47 and d18Oc data from seasonally aggregated proxy measurements¶

DEFINE UPDATING FUNCTIONS¶

Create updating function (Kalman filter)¶

Create function to track the statistics of the likelihood (combining just the reconstruction data)¶

EXECUTE UPDATING FUNCTIONS - SEASONAL¶

Update seasonal prior with aggregated data¶

Plot seasonal posterior in D47 domain¶

EXECUTE UPDATING FUNCTIONS - MONTHLY¶

Update monthly prior with aggregated data¶

Plot monthly posterior in D47 and d18Oc domain using aggregated data¶

PLOT PRIOR AND LIKELIHOOD FOR MANUSCRIPT¶

POSTERIOR - SEASONAL¶

Seasonal posterior in temperature and salinity domains (data aggregated per specimen and season)¶

Convert seasonal prior, likelihood and posterior to temperature and salinity and plot¶

Plot SAT and precipitation prior and posterior¶

Plot SAT, SST, SSS and precipitation posterior¶

Plot SAT, SST, SSS and precipitation posterior with uncertainties as 2 SD¶

POSTERIOR - MONTHLY¶

Monthly posterior in temperature and salinity domains with aggregated data¶

Convert monthly prior, likelihood and posterior to temperature and salinity and plot¶

Plot SAT and precipitation prior and posterior¶

Plot SAT, SST, SSS and precipitation posterior¶

Plot SAT, SST, SSS and precipitation posterior with uncertainty as 2 standard deviations¶